EVALUATING THE EFFECT OF COLOUR IN STEREOSCOPIC THREE-DIMENSIONAL PERCEPTION

Transcription

1 EVALUATING THE EFFECT OF COLOUR IN STEREOSCOPIC THREE-DIMENSIONAL PERCEPTION Shih-Chueh Kao Submitted in accordance with the requirements for the degree of Doctor of Philosophy The University of Leeds School of Design August 2015

2 The candidate confirms that the work submitted is his/her own, except where work which has formed part of jointly authored publications has been included. The contribution of the candidate and the other authors to this work has been explicitly indicated below. The candidate confirms that appropriate credit has been given within the thesis where reference has been made to the work of others. The work in Chapter 2 of the thesis has appeared in publication as follows: 'Scene Design for Stereoscopic 3D Perception in 3D Computer Animation', Proceeding of International Conference - Cinema, Art, Technology, Communication, 25 July 2010, Shih-Chueh Kao and Vanessa Walker I was responsible for literature review and research methodology. The contribution of the other author was conclusion and writing correction. This copy has been supplied on the understanding that it is copyright material and that no quotation from the thesis may be published without proper acknowledgement The University of Leeds and Shih-Chueh Kao The right of Shih-Chueh Kao to be identified as Author of this work has been asserted by him in accordance with Copyright, Designs and Patents Act ii

3 Acknowledgements This thesis is dedicated to my family. My wife, who has accompanied me throughout my study in Leeds, my parent and sister, for their support and understanding during my long term overseas study and work. I would like to thank those who have contributed to this study. My primary supervisor, Prof. Stephen Westland, for his advice and support throughout the entire thesis; my co-supervisors, Ms Vanessa Walker and Prof. Kenneth Hay, for their encouragement and mentoring; and a special thanks to Dr. Vien Cheung, for her guidance on many aspects that, looking back, are now so obvious. iii

4 Abstract An immersive stereoscopic three-dimensional (S3D) viewing experience is constructed on a successful scene design of the content. Colour is a remarkable factor that advantages human depth perception. Previous works in colour and depth perception have shown that long-wavelength stimuli, such as red or yellow, compete with short-wavelength stimuli such as blue or green, reds and yellows appear closer than blues or greens. Saturation and brightness are considered influential in spatial design. High saturated colours tend to capture more attention than neutral colours, and brighter colours appear advance while darker colours appear receded. This research explores stereoscopic depth perception based on investigating previous findings in colour and depth perception under a current 3D cinema viewing condition. The effect of colour on stereoscopic depth perception are examined from psychophysical and depth quality viewpoints. It considers the way in which depth perception is influenced by different decisions of hue, saturation and brightness, and identifies the thresholds and depth quality of stereoscopic depth perception in different colour arrangements. In particular, different levels of hue, saturation and brightness are tested on foreground and background colour palettes in computer-generated scenes. Psychophysical trials are utilised to examine the thresholds of observers depth perception. A depth quality assessment is then performed to evaluate different colour arrangements in practical scenes. A polarised projection system is built for stereoscopic viewing and the image stimuli are designed and rendered in the industry-standard 3D application package. The results indicate that hue, saturation and brightness are effective in stereoscopic depth perception. However, the discriminations of depth perception are equally efficient between the standard and reversed trials in hue and saturation section, which given an inference that contrast is a more dominant value in stereoscopic depth perception. Brightness is also a dominant factor but it can be resulted an undesired depth quality if the high brightness arranged in the foreground. iv

5 Table of Contents Acknowledgements... iii Abstract... iv Table of Contents... v List of Tables... ix List of Figures... xi Chapter 1 Introduction The Renaissance of Stereoscopic 3D Film Stereo Experience Improvement from Outset Does Colour Design affect S3D Aims of Research Thesis Overview Publications... 7 Chapter 2 Stereoscopic Vision and Colour in Depth Perception Depth Perception Psychological Depth Cues Physiological Depth Cues Cue Combinations Interaction between Monocular and Binocular cues Depth Perception in Computer-Generated Imaginary (CGI) Stereoscopic Vision Stereoscopic Display Auto-stereoscopic Display Stereoscopic 3D Film S3D Video Production Pipeline Colour Design in Filmmaking Colour Perception Colour Attributes Colour in Depth Perception Colour Contrast Contrast of Hue Contrast of Saturation Contrast of Value v

6 2.9 Colour Contrast and Depth Perception Depth Enhancement in Stereoscopic 3D Films Conclusion Chapter 3 Psychophysical Methods, Image Quality Assessment and Stereoscopic 3D Presentation Psychophysics The Psychometric Function Forced-Choice Procedure Method of Limits Method of Constant Stimuli The Method of Adjustment Adaptive Staircase Choice of Psychophysical Method Image Quality Assessment DSCQS Method Single-stimulus Method Stimulus-comparison Method Choice of Subjective Methods Experimental Conditions Statistical Analysis on Visual Experiment T-Test in Psychophysical Experiments and Visual Grading Univariate Analysis of Covariance Stereoscopic 3D Projection Polarised Projection Projectors Linear and Circular Light Polarisations Scene Design for Stereoscopic Presentation Foreground Mid-ground and Background Objects Scene Design for Comfort Viewing Stereoscopy in 3D Application Packages Stereoscopic Rendering HSV in Maya Conclusion vi

7 Chapter 4 Psychophysical Evaluation of Saturation, Hue and Brightness on Stereoscopic 3D Perception Introduction Hypothesess Methods Observers Stimuli Scene design Stimulus Images and Colour Values Apparatus Projection Validation and Issues of Applying in Stereoscopic Projection Procedure Results Saturation Hue Brightness Summary of Results Discrimination of stereoscopic depth changes in different colour arrangements Significance and Comparison of Thresholds Obtained Comparison of standard colour and reversed colour arrangements Chapter 5 Evaluating Saturation, Hue and Brightness in Stereoscopic 3D Perception by Image Quality Assessment in CG 3D Scenes Introduction Aims and Predictionss Methods Observers Procedure Scoring Scale Apparatus Stimuli Scene Design Colour Values Results vii

8 5.4.1 Saturation Hue Brightness Summary of Results Chapter 6 General Discussion Saturaion and Hue Controversy of Brightness The Dominance of Contrast Limitations Chapter 7 Conclusions Research Summary Possible Extensions and Future Works Confirmation of the Thesis Appendix A Image Stimuli for the Psychophysical Evaluation in Chapter Appendix B Instruction Sheets B.1 Instruction Sheet for Psychophysical evaluation B.2 Instruction Sheet for Depth Quality Assessment References viii

9 List of Tables Table 2.1 Depth cues in computer graphics Table 2.2 Advantages and disadvantages of current stereoscopic viewing techniques Table 2.3 Timeline of 3D Filmmaking Table 3.1 Five point scoring scale and associated terms in Singlestimulus method Table 3.2 Scoring scale and associated terms in Stimuluscomparison method Table 4.1 List of trials in the experiment Table 4.2 Colour values in standard trials Table 4.3 Psychophysical thresholds at 75% of psychometric function Table 4.4 T-test result concerning the standard and the reversed trials Table 4.5 Proportion of response in comparison stimulus Table 4.6 T-test results of slopes Table 4.7 Univariate analysis of variances Table 4.8 Psychophysical thresholds at 75% of psychometric function Table 4.9 T-test result concerning Standard and Reversed trials Table 4.10 Proportion of response in comparison stimulus Table 4.11 T-test results of slopes Table 4.12 Univariate analysis of variances Table Psychophysical thresholds at 75% of psychometric function Table 4.14 Proportion of response in comparison stimulus Table 4.15 T-test results of slopes Table 4.16 Univariate analysis of variances Table 4.17 Confirmation of research questions Table 5.1 Colour values in test images Table 5.2 Averaged scores from scene 1, scene 2 and scene 3 in the saturation section Table 5.3 T-test concerning means and the standard score Table 5.4 T-test concerning the standard and the reversed trials ix

10 Table 5.5 Averaged scores from scene 1, scene 2 and scene 3 in the hue section Table 5.6 T-test concerning means and the standard score Table 5.7 T-test concerning the standard and the reversed trials Table 5.8 Averaged scores from nineteen subjects in the brightness section Table 5.9 T-test concerning means and the standard score Table 5.10 T-test concerning the standard and the reversed trials Table 5.11 Means of score in 6 trials Table 5.12 Confirmation of the predictions Table A.1 Image stimuli in the saturation section Table A.2 Image stimuli in the saturation section Table A.3 Image stimuli in the saturation section x

11 List of Figures Figure 1.1 Growth of 3D Display Between 2008 to Figure 2.1 Depth Perception Cues... 9 Figure 2.2 Linear Perspective Figure 2.3. Atmospheric Perspective Figure 2.4 Texture Gradient Figure 2.5 Elevation Depth Perception Figure 2.6 Light and Shade Figure 2.7 Colour Depth Cue Example Figure 2.8 Overlapping Figure 2.9 Relative Size Figure 2.10 Motion Parallax Figure 2.11 Binocular Disparity Figure Convergence Figure Accommodation Figure 2.14 Just-discriminable ordinal depth thresholds Figure 2.15 Horopter and fusion area Figure 2.16 Liquid crystal display (LCD) shuttering unit and shuttering glasses Figure 2.17 Anaglyph glasses Figure 2.18 Polarised glasses and projection Figure 2.19 Principle of Lenticular Technique Figure 2.20 Current production pipeline in use at DreamWorks Animation Figure 2.21 Embed Stereoscopy Production Pipeline Figure 2.22 Colour design in the film The Sixth Sense Figure 2.23 Colour design in the film The Matrix Figure 2.24 Colour design in the film Prometheus Figure 2.25 Chromatic Aberration Figure 2.26 Colour wheel of contrast of hue Figure 2.27 Example of saturation contrast Figure 2.28 Example of value contrast Figure 2.29 Depth enhancement in film Monsters vs. Aliens Figure 2.30 Depth enhancement in film Friday the 13 th Part III xi

12 Figure 2.31 Depth enhancement in film Coraline Figure 2.32 Depth enhancement in film The Hobbit: An Unexpected Journey Figure 2.33 Depth enhancement in film Avatar Figure 2.34 Depth enhancement in film Life of Pi Figure 3.1 Example of psychometric functional sigmoid curve Figure 3.2 Example of staircase curve Figure 3.3 Presentation structure in DSCQS method Figure 3.4 A five point scoring sheet of DSCQS method Figure 3.5 Polarised Projection Technique Figure 3.6 Foreground, mid-ground and background areas in scene design Figure 3.7 Comfort stereoscopic viewing range Figure 3.8 Six parameters that construct viewing camera frustum in Autodesk Maya Figure 3.9 Stereo camera attributes in Autodesk Maya Figure 3.10 Images rendered from stereo camera rig in Autodesk Maya Figure 4.1 Stimulus scene design for a comfortable stereoscopic viewing Figure 4.2 Dimensions of foreground colour palettes, background colour palettes and canvas Figure 4.3 HSV colour penal in Autodesk Maya Figure 4.4 The interface of WAT-C colour contrast analyser Figure 4.5 The width of actual projection size Figure 4.6 The parallax of actual projection Figure 4.7 Polarised projection utilised in this work Figure 4.8 Polarised projection and viewing distance Figure 4.9 Image used to align left and right projections Figure 4.10 Aligned projection Figure 4.11 Equipment arrangement for measurements Figure 4.12 Efficiency of light transmission Figure 4.13 Transition between trials Figure 4.14 Psychometric function of the standard trial Figure 4.15 Psychometric function of the reversed trial Figure 4.16 Linear regression of the standard trial xii

13 Figure 4.17 Linear regression of the reversed trial Figure 4.18 Psychometric functions of the standard trial Figure 4.19 Psychometric functions of the reversed trial Figure 4.20 Linear regression of the standard trial Figure 4.21 Linear regression of the reversed trial Figure 4.22 Psychometric functions of the standard trial Figure 4.23 Psychometric functions of the reversed trial Figure 4.24 Linear regression of the standard trial Figure 4.25 Linear regression of the reversed trial Figure 4.26 Psychometric functions comparison Figure 5.1 Illustration of gray interval between trials Figure 5.2 Nine points scoring scale based on ITU-R BT Figure 5.3 Scene Figure 5.4 Scene Figure 5.5 Scene Figure 5.6 Colour wheels with hue values and cool-warm sites Figure 5.7 Reference images for scene 1, scene 2 and scene Figure 5.8 Box plot in the saturation section Figure 5.9 Box plot in hue section Figure 5.10 Box plot in the brightness section xiii

14 xiv

15 1.1 the renaissance of stereoscopic 3D film CHAPTER 1 Introduction 1.1 The Renaissance of Stereoscopic 3D Film The stereoscopic three-dimensional (S3D) cinema was once at its peak in the 1950s. Duncan (2006) described that more than 65 stereoscopic feature films were released by Hollywood from 1952 to However, the popularity of the medium rapidly decreased due to technical problems such as the film projection glitch, images misalignment, and visual discomfort. In 2005, RealD 1 developed the single digital S3D projection system and Lipton (2007) pointed that the technique solved most of those technical problems and stimulated the recent wave of producing and projecting films in the S3D format. Digdia (2011), the new 3D entertainment report, indicated that the number of 3D movie screen worldwide has grown from 5000 in the middle of 2009 to 22,500 in the end of 2010, which is an increase of 450% in 18 months. The report also showed an average of 69% of world s modern screens are now converted into digital 3D. The worldwide total for digital 3D screens reached 45,545 in 2012 marking a 26.5% increase from the 36,000 installed at end Hsieh (2012) indicated that the growth of 3D display in home entertainment, cameras and mobile devices would reach 226 million units in 2019 (see Figure 1.1). Quixel Research (2009) reported that 46% of those surveyed want to receive 3D content via their cable or satellite providers and 30% of the participants would be interested in changing their content provider in order to receive 3D content. 1 RealD is a company that develops 3D cinema technology. 1

16 1.2 Stereo Experience Improvement from the Outset Figure 1.1 Growth of 3D Display Between 2008 to D display products are expected to achieve 226 million productions in The figure is adapted from 3D Display Technology and Market Forecast Report by Hsieh (2012). However, Shibata et al. (2011) indicated that the S3D entertainment s resurgence in popularity has been accompanied by serious concerns about adverse effects regarding the S3D viewing discomfort. Hoffman (2008) and Lang (2010) et al. concluded that the practical construction of S3D content that leads to a comfortable and immersive viewing experience remains to be a great research challenge due to the complexity of the human visual system. 1.2 Stereo Experience Improvement from the Outset This wave of S3D has encouraged new techniques to improve stereo viewing experience throughout entertainment industry. 3D Focus (2013) reported that Disney is beginning a test on a trifocal camera system that comprises an Arri 2 M camera with two micro HD cameras. The dual micro cameras capture depth information and then are analysed by stereoscopic 2 A supplier of motion picture film equipment. 2

17 1.2 Stereo Experience Improvement from the Outset software for post-production in 3D. The new designed S3D camera would avoid the need to carry heavy 3D rig camera system and deliver a 3D experience without lens misalignment. SensoMotoric Instruments 3 have developed world s first full eye tracking and Microsoft Kinect 4 compatible stereoscopic eyewear that allows people to interact with their environments through gaze. RealD has announced Precision White Screen technology that combines 2D white screen performance and the ability to project polarised 3D films. The new screen provides 40% brighter projection than a traditional silver screen. Technologies described above improved audience s 3D viewing experience. However, a successful stereoscopic 3D viewing not only relies on advanced technologies but also is based on creative stereoscopic content design. Scene design is important for S3D production as the creations in this phase directly affect depth performance of the content. Well-designed stereoscopic scenes strengthen the integration of storytelling and immersive experience. The key factors of successful stereoscopic illusion are no longer only a afterthought process in post-production but are moving further upstream in stereographers 5 depth script and computer-generated imagery (CGI) artists comprehensive contemplation. Robertson (2008) described the filmmaking process is being rethought and turned from a linear workflow into a more parallel and interactive procedure. When filmmakers consider stereoscopy from its outset, they must be aware of the importance of such production pipeline change. Stereographers play an important role since such change because their creations are the foundation for the subsequent stereo effect procession. Scene design elements that relative to depth perception such as perspective, object shape, colour and camera movement, must be taken into 3 A company that develops solutions for gaze and eye tracking system. 4 A motion sensing device developed by Microsoft for video game console. 5 Professionals in creating stereoscopic 3D perception. 3

18 1.3 Colour Design and S3D account in the process of S3D filmmaking. Moreover, with much more software being retooled for enabling S3D moviemaking, CGI artists hold further power to control depth perception and decrease problematic stereo illusion. Numerous technical studies have discussed stereo improvements from the aspect of post-production and display viewing discomfort, such as converting issues in 3D videos post-production and processing (Smolic, Kauff, Knorr and Hornung, 2011) and 3D TV viewing discomfort and assessment methods of 3D viewing evaluation (Lambooij and Heynderickx, 2011). These works mainly focused on hardware improvements instead of the upstream stage of scene design. The very initial stage for stereoscopy deserves more attention and the research in content design may reveal a new chapter in S3D viewing improvement. This thesis investigated S3D viewing from a visual perception point of view and addressed on the evaluation of viewing content. Colour is a critical element in content design and visual perception, and its potential influence in S3D viewing is what this study mainly focused. 1.3 Colour Design and S3D S3D viewing is based on the principle of binocular disparity. Dresp and Guibal (2004) indicated that colour is a remarkable monocular depth cue that advantages binocular depth performance. Hue, saturation and brightness are crucial considerations where depth is most concerned. Brewster s (1851) work in colour stereopsis has shown that long-wavelength stimuli, such as red or yellow, compete with short-wavelength stimuli such as blue or green, when viewed binocularly, reds and yellows appear closer than blues or greens. In other words, warmer hues associated with longer wavelength help objects standout; cooler hue associated with shorter wavelength help objects recede. Saturation and brightness are considered influential in spatial design. Mount, Case, Sanderson and Brenner (1956) and Egusa (1983) concluded that high 4

19 1.4 Aims of Research saturated colours tend to capture attention before their counterparts because of their greater strength. This leads the viewer to conclude that a more saturated colour advances while neutral colours tend to recede. If two or more adjacent colours are unequal in brightness, only the lighter colour will appear closer while the darker colour will recede. This thesis aimed to find out how monocular depth cue such as colour, and binocular depth cue influenced depth performance in computer-generated 3D scenes. This is a relevant topic for depth perception given the results of past works, such as Luckiesh (1918), Mount (1956), Sundet (1978) and Egusa (1983), that indicated colour is an influential depth cue over human depth perception. In addition, the results from this study will provide an empirical evidence to support a proper colour design of stereoscopic 3D content that applicable to fields such as 3D filmmaking, animation, gaming and virtual simulation. Furthermore, the thesis investigated the effect of colour in stereoscopic 3D perception based on a 3D cinema simulation while few attempts had looked the topic from a cinematographic design aspect and done by a polarised 3D projection. 1.4 Aims of Research To validate this thesis, it is necessary to Investigate human depth perception and S3D content production and highlight the necessity of considering stereoscopy from the outset in scene design, and then designate conceivable influence of colour perception on stereoscopic depth viewing. Build a polarised projection system that is based on the current format of 3D cinema to view S3D stimuli. Produce S3D content for stereoscopic depth perception evaluation. Experimentally evaluate colour considering hue, saturation and brightness on stereoscopic depth perception through examining psychophysical thresholds. 5

20 1.5 Thesis Overview Experimentally assess stereoscopic depth quality considering hue, saturation and brightness in detailed 3D scenes based on international standard procedure. Conclude the comparison of colour arrangements for S3D content productions. 1.5 Thesis Overview This thesis is organised as follows: Chapter 2 reviews literatures of depth perception, stereoscopic vision and related works that investigated colour in depth perception. This chapter also introduces S3D filmmaking pipeline and current S3D viewing technologies. Chapter 3 describes general methods of psychophysical experiment and depth quality assessment. Furthermore, polarised projection techniques and S3D image production are discussed in this chapter. Chapter 4 presents psychophysical experiments to evaluate S3D depth perception by setting different levels of hue, saturation and brightness in colour palettes stimuli. Psychophysical functions and Statistical data are analysed in this chapter. Chapter 5 presents depth quality assessment based on ITU- Recommendation to evaluate S3D depth quality by hue, saturation and brightness in computer-generated imagery S3D scenes. Statistical data are analysed and the results are compared with the results from Chapter 4. Chapter 6 discusses general conclusions and validation to related works and Chapter 7 summarises research results and indicated possible extensions. 6

21 1.6 Publications 1.6 Publications The publications generated from this work are as follows. Kao S. (2014) Stereoscopic Three-Dimensional Depth Quality Assessment in Different Colour Arrangements, European Conference on Visual Perception 2014, Belgrade, Serbia Kao S. (2013) Brightness Contrast in Stereoscopic 3D Perception, Proceedings of Multimedia, Interaction Design and Innovation 2013 Conference, Warsaw, Poland Kao S. & Westland S. (2012) Colour Hue Adjustments for Stereoscopic 3D Perception, International Colour Association 2012 Conference, Taipei, Taiwan Kao S. (2011), The Adjustments of Colour Saturation for Stereoscopic 3D Perception, Digital Future 2011: Stereoscopic Imaging Conference, London, UK Kao S. & Walker V. (2010), Scene Design for Stereoscopic 3D Perception in 3D Computer Animation, Proceeding of International Conference - Cinema, Art, Technology, Communication, Avanca, Portugal Kao S. (2009), Consider Stereoscopy from Outset: The Adjustments in 3D Scenes for Stereoscopic Presentation, SMPTE Annual Tech Conference & Expo 2009, Hollywood, USA 5

22 2.1 Depth Perception CHAPTER 2 Stereoscopic Vision and Colour in Depth Perception As the stereoscopic illusion is generated by depth perception from human visual system, this chapter will review the theory of depth perception, stereoscopic vision and then discuss related works in depth enhancement by colour adjustments. 2.1 Depth Perception When one object appears to overlap another, we assume that the overlapping figure is closer to us. When an object appears to be smaller than its expected or known size, we presume that it is far away. Human sense of space depends on comparison of objects that share a linear perspective 6. Okoshi (1976) defined depth perception as the human visual system s mechanism that relies on both physiological and psychological cues to make accurate estimates of the depths of objects. He explained that depth cues are the clues that give the human brain information of the orientation of objects in the perceived image, and permit the accurate perception of the image to be interpreted by the human brain. Chen (2005) stated the psychological issues about human depth cue perception, as including linear perspective, atmospheric perspective, texture gradient, elevation, light and shade, colour, overlapping, relative size and motion parallax ; the physiological depth cues including binocular disparity, convergence and accommodation (see Figure 2.1). 6 Linear perspective is a depth cue that is related to both relative size and the next depth cue. 8

23 2.1 Depth Perception Figure 2.1 Depth Perception Cues - Depth perception cues categorised by Chen (2005) Psychological Depth Cues Chen (2005) explained that psychological depth cues are based on the interpretation and analysis of the retinal image that is caused by the working of the visual cortex 7 in the brain. Figures 2.2 to 2.10 illustrate and describe 7 The visual cortex of the brain is the part of the cerebral cortex responsible for processing visual information. 9

24 2.1 Depth Perception the following different types of psychological cues: linear perspective; atmospheric perspective; texture gradient; elevation depth perception; light and shade; colour; overlapping; relative size; motion parallax. Figure 2.2 Linear Perspective Objects appear farther are smaller than objects that are closer. The sizes changed in inverse proportion to their depth. Figure adapted from Chen (2005). Figure 2.3. Atmospheric Perspective Due to the atmospheric scattering of light, farther objects often appear hazy or bluish. Figure adapted from Chen (2005). 10

25 2.1 Depth Perception Figure 2.4 Texture Gradient - Textures on an object appear coarse if the object is close to the viewer, and fine if the object is far from the viewer. Figure adapted from Bradley s (2012) 11 ways to add depth to a design [Online]. Available: February 2013] Figure 2.5 Elevation Depth Perception - Objects lower in the visual field and have more distance from the horizon line are perceived closer than objects higher in the visual field and have less distance to the horizon line. Figure adapted from Chen (2005). 11

26 2.1 Depth Perception Figure 2.6 Light and Shade The surface looked brighter when the object is near the light source. The left ball looked flat due to the light source is farther away. The farther parts of an object's surface are from the source of light, the more shadowed and less bright they will appear. Varying shading and lighting then provide information about distances of objects from the source of light, and may serve as a cue to the distance of the object from the observer. Figure adapted from Chen (2005). Figure 2.7 Colour Depth Cue Example, Van Gogh, The Plain of Auvers, Carnegie Museum of Art - Some colours appear to stand forward, and some to recede, the softening of aerial perspective may play a role. In the painting The Plain of Auvers by Van Gogh, the sky appears to stand a good several inches behind the horizon, the plain looks gradational by different level of green colours. Figure adapted from Van Gogh (1890). 12

27 2.1 Depth Perception Figure 2.8 Overlapping In the real world, the fully visible objects are perceived closer than partially blocked objects. Figure adapted from Chen (2005). Figure 2.9 Relative Size The size of the object becomes smaller as an object's distance from the viewer increases. Generally, relative larger objects are perceived closer than are smaller objects. Figure adapted from Chen (2005). 13

28 2.1 Depth Perception Figure 2.10 Motion Parallax - Objects at different distances associate to the observer are perceived as moving at different speeds. This is due to human eye movement that associates to the spatial environment. Objects move at different rates at varying distances relative to their position and distance from the observer. Figure adapted from Ware (1999) Physiological Depth Cues Unlike psychological depth cues that achieve depth perception with cues pertaining to the visual pattern, physiological depth cues are formed by muscle adjustment of the eye or head movement and require both eyes to create the depth. Figures 2.11 to 2.13 describe binocular disparity, convergence and accommodation respectively. Figure 2.11 Binocular Disparity - Foley et al. (1997) considered binocular disparity is the most important depth perception cue over medium viewing distances. When looking at an object, human eyes see different images. The difference between the images of the same object projected onto each retina is binocular disparity. The level of the depth depends on the parallax that is the angle formed by the optical axes of each eye converging on an object. The circle line that can be drawn through all points in space that stimulate corresponding retinal points for a given degree of convergence is called the horopter. See more description about binocular disparity in Section 2.2. Figure adapted from Foley et al. (1997). 14

29 2.1 Depth Perception Figure Convergence - The distance between eyes is fixed 6.5cm but when the eyes focus on an object, the angle between the optical axes changes. When looking at the farther cone the eye orientation is nearly the same but when look at the closer cone, the left eye looks more to the right and the right eye looks more to the left. The angle between the optical axes of both eyes is convergence, which gives our brain a hint about the distance to an object. Figure adapted from Hurvich (1981). Figure Accommodation Hurvich (1981) indicated that accommodation is occurred when the eye lens change their curvature to associate different distance of objects to form sharp retinal images. To focus on far objects the lens becomes relatively flat and to focus on nearer objects the lens becomes more curved. Figure adapted from Hurvich (1981). 15

30 2.1 Depth Perception Cue Combinations Psychological and physiological depth cues are both important in achieving depth perception. However, these depth cues normally don t function alone as our world contains complex factors to form visual perception. Howard and Rogers (1995) indicated that cues may function in harmony or can be in conflict. There are several ways that different sources of information may interact described as follows. Cue averaging. Cue combination is based on independent cues that are linearly combined with different weights assigned to each cues. Frisby, Buckley and Freeman (1996) demonstrated this form of interaction experimentally. Cue dominance. Perception judgement is based on the dominant cue, which the other cue is restrained in conflict. Screen edge effect is an example of the situation. The occlusion from the screen border dominates the depth perceiving and breaks completed visual perceiving of objects in the screen. Braunstein (1976) found that when velocity gradients 8 were placed in conflict with texture gradients in perception of slant, velocity dominated texture. Cue reinterpretation. After combination some cues may be affected to be compatible with other cues and interpreted differently. An instance is the kinetic depth effect. When the silhouette of a rotating object, such as a bent piece of wire, appears three-dimensional even without the disparity cue and appears two-dimensional when the motion stops. Normally, the more consistent the cues in visual perception the more accurate the depth perception. Cutting & Vishton (1995) and Nagata (1993) discussed the relative information strength of depth cues at various 8 Velocity gradient is the rate of change of velocity of propagation with distance normal to the direction of flow. 16

31 2.1 Depth Perception distances. Each cue by itself is an unclear reference of distance and layout. However, when combining several cues this ambiguity may be reduced, see Figure 2.14 for the depth thresholds from the psychophysical experiment done by Nagata (1993). Figure 2.14 Just-discriminable ordinal depth thresholds The thresholds were studied by Nagata (1993) and it indicates a function of the logarithm of distance from the observer. More potent sources of information are associated with smaller depth-discrimination thresholds. Cutting and Vishton (1995) then adapted this figure to study perceiving layout and distance. Figure adapted from Cutting and Vishton (1995) Interaction between Monocular and Binocular Cues Human eyes perceive monocular depth cues from objective stimulus that presents stimulus image on the retina and the effects include light, occlusion, objects size, height in the visual field and relative motion gradient or parallax. These cues require only one eye to perceive. In contrast, binocular depth cues are available when both eyes are used to view a scene due to binocular depth is triggered by the optics and muscles of the eye. The 17

32 2.1 Depth Perception effects include accommodation, collimation, convergence and binocular stereopsis. Many works have investigated the performance of spatial perception from strengths of monocular and binocular depth cues. Cutting and Bruno (1988) concluded that depth cues can be either additive, which depth perception is additive to cues employed, or non-additive, which the result can be superadditives or subtractive. Tittle and Braunstein (1993) asked subjects to recover the perceived structure of a rotating cylinder and found that motion had a super-additive effect on stereoscopic viewing. Van der Meer (1979) also investigated the effect of binocular disparity and perspective on depth perception and found that when the perspective and binocular cues were presented in conflict with each other, most observers used stereopsis over perspective. He then concluded that depth perception increased when these cues presented congruent information. Moreover, the interaction of perspective and stereopsis were concluded to be additive. Braunstein (1986) found that linear perspective cue and stereopsis were additive in terms of the perception of slant. The result is consistent with Van der Meer s (1979) conclusion. Studies investigating stereopsis as a depth cue concluded that stereopsis has positive interaction with binocular depth to form more accurate depth perception. Kim et al. (1987) asked subjects to track an object manually in a 3D scene to examine the effect of binocular disparity and perspective cues. The results indicated that stereopsis improved the overall tracking performance. Yeh and Silverstein (1992) examined subjects reaction time of spatial judgement based on the effect of stereopsis. They concluded that stereopsis improved the performance for both depth and altitude judgements in 3D space and significantly decreased the reaction time of making these judgements. Howard and Rogers (1995) claimed the only monocular cue that provides a similar degree of depth resolution created by binocular cue is motion parallax. Monocular motion parallax and binocular disparity are closely related. Rogers and Graham (1982) demonstrated temporally 18

33 2.1 Depth Perception separated successive views could provide the same information to the visual system as spatially separated views Depth Perception in Computer-Generated Imaginary (CGI) Monocular depth cues are generally available in CGI recreation and replication. Hendrix and Barfield (1995) indicated binocular disparity is also available in computer-generated environment for three-dimensional design. He concluded the variables for both monocular and binocular depth cues in relation to the computer graphics in Table 2.1. All monocular depth cues can be recreated in CG environment but it hasn t been concluded that using current technology can reproduce all binocular depth cues. 19

34 2.1 Depth Perception Table 2.1 Depth cues in computer graphics - Hendrix and Barfield (1995) concluded depth cues that can be manipulated by computer graphic technology. This table is edited from Hendrix and Barfield (1995). Monocular depth cues Binocular depth cues Type of Depth Cue Luminance/brightness Ambient hue Shadows/highlights Colour (hue, saturation, brightness) Perspective, compression and density gradients Linear perspective Size distance invariance Size by occlusion Familiar size Height in the visual field Motion parallax Kinetic depth effect Accommodation Convergence Disparity Computer Graphics Feature Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes No No Yes Hendrix and Barfield (1995) concluded that monocular depth cues are helpful in receiving binocular depth cues. In the field of computer graphics, CGI artists can adjust monocular cues to enhance stereo cues. Salzmann et al. (2009) indicated that perceiving an object in a virtual world is much less precise than perceiving in the real world. This point of view discords with Boyd s (2005) conclusion that depth cues in the virtual and real worlds are compatible. He concluded that the real world s depth cue is referential when creating depth cue in virtual world. However, the designer must consider what is most necessary to present in scenes due to limited computer resources. It is certain that settings of models play a decisive role of 20

35 2.2 Stereoscopic Vision determining depth cue perception and of a key adjustment to balance the quality and computing resource. Based on the conclusion of Hendrix and Barfield (1995), this thesis created stimulus images in CGI environment to evaluate stereoscopic depth perception. 2.2 Stereoscopic Vision This section will focus on stereoscopic vision triggered by binocular perception mentioned in the previous section. Bruce, Green and Georgeson (1996) claimed that stereoscopic 3D vision is reconstructed by the brain through associating visual information with information from other modalities. The two eyes are separated and see the world from slightly different viewing points, which result two different retinal projections of an object in space. Wheatstone (1838) discovered stereoscopy and invented stereoscope 9, which contributed critically to human depth perception. His discovery showed that the brain uses binocular disparity to estimate the relative depths of objects in the world. This process is called stereopsis. Viewers will see two superimposed images that appear to be horizontally displaced to slightly different degree when looking at stereoscopic displays without the appropriate goggles such as polarisers, anaglyph glasses or shutter glasses. This displacement between the left and right images is called parallax. When the two images are shown separately to each eye simultaneously, parallax produces a retinal disparity and results the depth perception. In Palmer s (1999) experiment, stereoscopic vision is considered remarkably good at regions close to the horopter. Panum (1858) defined that horopter is the line that passes through all points in space that corresponding retinal points for a given degree of convergence. The theoretical horopter is a circle, known as the Vieth-Müller Circle, which passes through the fixation point and the nodal points of both eyes (see Figure 2.15). Palmer s (1999) 9 Stereoscope is a technique for creating or enhancing the illusion of depth in an image by means of stereopsis for binocular vision. 21

36 2.2 Stereoscopic Vision experiment showed that the empirical horopter lies slightly behind the theoretical horopter. However, the difference between the theoretical horopter and the empirical horopter are very small and normally is ignored for practical purposes. Retinal disparity happens on the points that are not on the horopter line. Disparities behind the horopter are uncrossed and crossed when disparities in front of the horopter. The fused disparities within the region around the horopter line form an specific area called Panum s fusional area that was measured by Panum (1858) and translated into English by Hübscher (1940). Figure 2.15 illustrates a fixed area of a given retinal region; however, the actual size and shape of Panum s fusional area changes depend on the stimulus spatial characteristics. A phenomenon called diplopia happens when the disparity is large and the viewer sees double images. For example, when hold one finger up in front of the eyes and the other finger up at an arm s length, the close finger will appear double when the viewer focuses on the farther finger. At which fusion the largest disparity can occur is dependent several factors. Duwaer & van den Brink (1981) showed that the threshold of diplopia is based on the subject tested, subject training before the experiment, the criterion, and the conspicuousness of the disparity. 22

37 2.2 Stereoscopic Vision Figure 2.15 Horopter and fusion area. Points within Panum s area are fused into a single image. Points that are closer or farther produce double images of crossed or uncrossed disparity. Figure adapted from Palmer (1999). Limits in perceiving binocular fusional disparity discussed by Patterson and Martin (1992) indicated that larger size stimuli, such as +/-20 minutes of arc 10, has two times foveal region for disparity fusion than small sized stimuli, such as smaller than 15 minutes of arc. However, Howard and Rogers (1995) indicated up to 2 degrees of overall disparity between two images is tolerated before the sensation of depth is lost. Therefore, most individuals have an appreciation of depth beyond the diplopia threshold, the region where single vision has been lost. Woo and Sillanpaa (1979) reported 10 Minutes of Arc (MOA) is a unit of visual angular measurement, equal to one sixtieth (1/60) of one degree. 23

38 2.3 Stereoscopic Display crossed disparity has better stereoacuity 11 performance than uncrossed disparities. 2.3 Stereoscopic Display The principle of the stereoscopic display is to display the left view to the left eye, and the right view to the right eye. There are several ways to achieve this. Okoshi (1980), Pastoor and Wöpking (1997), and Sexton and Surman (1999) all noted that stereoscopic displays can be categorised based on the techniques used to display the right and left images be seen by the appropriate eyes. The current prevailing approaches of stereoscopic viewing require a viewing aid, such as glasses, to separate the right and left eye images. Stereoscopic displays that do not require viewing aid are known as autostereoscopic displays. The technique embeds the eye-addressing techniques in the display itself. Current developing autostereoscopic display technique is allow to have more than one viewer, which means it allows for more than one geometrically correct viewpoint. For stereoscopic displays that require viewing aid, the technique can be categorised into time-parallel and time-multiplexed. Time-parallel technique appearing both left and right eye views simultaneously. Time-multiplexed technique showing left and right eye views in rapid alternation and synchronised with a liquid crystal shuttering 12 system which opens in turn for one eye, while occluding the other eye. See Figure 2.16 for liquid crystal shuttering equipments. Pastoor and Wöpking (1997) indicated human visual system is capable of integrating the constituents of a stereo pair across a time lag of up to 50 ms 13 when viewing time-multiplexed system. The display 11 Stereoacuity is the smallest detectable depth difference that can be seen in binocular vision. 12 A Liquid Crystal Shutter is simply an LCD that has a single large pixel that covers the entire display area. The shutter is simply "open", or "closed". The display can be toggled between its open and closed state by applying a simple square wave drive voltage. 13 MS (millisecond) is a thousandth ( 1 / 1,000 ) of a second. 24

39 2.3 Stereoscopic Display will need to run at twice the frame rate 14 of the original sequence because of the rapid alternation of right and left eye images. For example, to display a 60 Hz sequence a 120 Hz display is required. If the frame rate is not fast enough, cross-talk may occur due to the persistence of CRT 15 phosphors. Cross-talk is an imperfect separation of the left and right eye views which can be perceptually annoying. Figure 2.16 Liquid crystal display (LCD) shuttering unit and shuttering glasses. Figure adapted from Chen (2005). There are several multiplexing methods for time-parallel stereoscopic displays including colour and polarisation. The colour method (producing what is known as an anaglyph), filters the left and right eye images with complementary colours such as red and green. To separate left and right images, the observer is required to wear anaglyph glasses. The limitation of this technique is that colour information is lost and only limited colour is possible through the binocular colour mixture. This well-known and inexpensive method has been used for stereoscopic 3D cinema and television, and is still popular for viewing stereoscopic images in print. See Figure 2.17 for anaglyph glasses. 14 Frame rate is the frequency (rate) at which an imaging device produces unique consecutive images called frames. 15 The cathode ray tube (CRT) is a vacuum tube containing one or more electron guns and a fluorescent screen used to view images. 25

40 2.3 Stereoscopic Display Figure 2.17 Anaglyph glasses -Complementary colours applied in the lens are normally red and green in Europe, and red and blue in USA. Figure adapted from Tague (2012) interstellardesign.com[ [Online]. Available: February 2013] Polarisation-multiplexed displays polarising light to separate left and right eye images. Left and right output devices, such as projectors, project light filtered by polarising lens or orthogonally oriented filters, using either linear or circular polarisation. The observers are required to wear polarised glasses to separate the left and right views. Pastoor and Wöpking (1997) indicated the technique provides preferred quality stereoscopic imagery without losing colour informationn and resolution. The system is the prevailingly used in current stereoscopic 3D cinemas. RealD, IMAX 3D and Dolby 3D 16 cinemas all adapt polarising projection system. The loss of lightness is the most significant issue of this system. Moreover, under linear polarised projection, cross-talk will occur if the viewer tilting head into incorrect position. See Figure 2.18 for an example of polarised glasses and projection. Table 2.2 summaries the advantages and disadvantages of current stereoscopic viewing techniques. 16 IMAX 3D and Dolby 3D are current prevailing 3D cinema technologies like RealD. 26

41 2.3 Stereoscopic Display Figure 2.18 Polarised glasses and projection. Figure adapted from Chen (2005). Table 2.2 Advantages and disadvantages of current stereoscopic viewing techniques. Techniques Advantages Disadvantages Incapable of multiviewer; Incompatible High entertaining for video LC Interlace Shutter game; middle cost. with LCD or progressive scan monitor. Anaglyph Polarisation Most economical, simple to view, need anaglyph glass only. Colour information of stereo illusion is more accurate than interlace and anaglyph presentation; low-priced glass Inaccurate colour information; serious ghosting problem. Expensive projection system; very technical Auto-stereoscopic Display As mentioned in the previous section, auto-stereoscopic displays do not require viewers to wear viewing aids. The technique is based on direction multiplexing through different two-dimensional images projected across the viewing field. Pastoor and Wöpking (1997) noted that auto-stereoscopic techniques apply optical principles such as reflection, diffraction, occlusion and refraction to direct the light to the appropriate eye to form different perspective views. Sexton and Surman (1999) mentioned that the most 27

42 2.3 Stereoscopic Display prevailing auto-stereoscopic techniques are based on the principle of parallax barriers and lenticular arrays. The lenticular technique is based on refraction, and places an array of vertical cylindrical lenses in front of columns of pixels to represent parts of the left- and right-eye views. The refraction directs the light of image points emitted in specific directions in the horizontal plane. The well-known example of this technique is 3D postcards; the card is covered by a convoluted plastic sheet on the surface. The parallax barrier technique is based on occlusion and part of the image can only be seen by one eye and is invisible to the other eye (see Figure 2.19). The viewer is required to view the image from a certain distance and angle; one eye can only see the appropriate view, as the other view is occluded by the barrier effect of the vertical slits. Both lenticular and parallax barrier viewings require precise alignment of the picture splitter with the vertical left-right image strips. Current auto-stereoscopic displays all require a very stable position of the picture elements. Figure 2.19 Principle of Lenticular Technique. Figure adapted from Gondane (2014) Lenticular blog [Online]. Available: February 2013] 28

43 2.4 Stereoscopic 3D Film 2.4 Stereoscopic 3D Film Stereoscopy has been embraced by many fields and filmmaking industry is the major medium that introduced the effect to people. Hayes (1998) described the rise, fall and second rise of commercial Hollywood 3D films in terms of the first rise during and the second boom during However, Zone (2007) claimed the periods of the 3D film industry should encompass longer spans of time. The Novelty Period, , covered the invention of filmmaking and the later efforts to include sound and colour. Short and experimental films of this period typically included the use of gimmicks, include 3D effects that leap off the screen. During this period, tests and investigations dominated the primarily short stereo films. From Niagara Falls, a silent 3D documentary directed by Porter (1915) to the artistic animation of Oskar Fischinger (1935) in the early 1930s, the film industry experimented with this new form of expression. The second period, The Era of Convergence, , began with Hollywood s venture into 3D feature films with the first American commercial film Bwana Devil released in Zone (2007) defined this period includes The Golden Age of Hollywood 3D Films like House of Wax released in 1953 and Creature from the Black Lagoon released in 1954 demonstrated the advancements of the technology of 3D film production, but also revealed the exploitation of the medium. The Era of Convergence also covers the rise in production of 3D pornographic films such as Bellboy and the Playgirls released in 1962 and The Stewardesses released in Other films in this era include Jaws 3D and Amityville 3D in Hayes (1998) described these films contained numerous in-your-eye types shots which arranged objects in the foreground to standout from the screen and pleased audiences. Zone (2007) concluded that although Hollywood had temporarily lost interest in 3D films in this era, specialised venues such as theme parks and large-format theatres would begin production and eventually monopolise the stereo production industry. 29

44 2.4 Stereoscopic 3D Film The Immersive Era, 1985-present, began with the first IMAX 3D film We Are Born of Stars 3D as well as with Captain EO in 1985, the first 3D film produced specifically for a theme park attraction. Zone (2007) included 3D for special venues such as theme parks and IMAX theatres, digital 3D, virtual reality, and 3D for the home in this period. See Table 2.3 for a timeline of 3D filmmaking. Table 2.3 Timeline of 3D Filmmaking Timeline Name Description present The Novelty Period The Era of Convergence The Immersive Era Invention of film, experiments with production, sound and colour. Gimmicky 3D that leaps off the screen. Hollywood s venture into feature-length 3D starting with Bwana Devil in Includes dual-strip 3D film of the 1950s through the single-strip 3D films of the 1970s and 1980s. Begins with first IMAX 3D film Transitions and Captain EO in 1985, the first 3D film made for theme park attraction. Includes 3D for special venues, digital 3D, virtual reality and 3D viewed or produced in the home S3D Video Production Pipeline In the past few years, digital technologies have blurred the lines between many parts of digital filmmaking. The renown example is from the director of the S3D film Avatar, Cameron (2009) presented a filmmaking that over 60% was done by CGI and brought the notion of a production pipeline that 30

45 2.4 Stereoscopic 3D Film considering stereoscopic 3D viewing experience. This section will describe the change in pipelines of digital filmmaking and indicate the significance of scene adjustment. A clear pipeline of production is helpful in understanding the value and the goal of each phase. As the structure of the pipeline changes, the procedure demanded in tandem and parallelism has become essential. Wallen (2009) described serialised tasks made at DreamWorks Animation; these start with storyboards, bring the work into layout, at the same time, other departments begin working on modeling, surface textures and then backgrounds. The key is performing these works in parallel, such as animation and lighting, will migrate to a fully parallel workflow, which allows departments at different stages of the pipeline to view and collaborate. (see Figure 2.20). Communication cycles occur between movie director and departments. The department makes changes, the director gives his/her opinion on it and the artist makes the changes based on those directions, the same cycle keep repeating until it is right. 31

46 2.4 Stereoscopic 3D Film Figure 2.20 Current production pipeline in use at DreamWorks Animation. Figure adapted from Ramanujam (2010), Intel Blog, Issue 2, article 'DearmWork Animation and Intel' [Online]. Available: October 2011] However, producing stereoscopic content differs from traditional flat-field movies. It s not just adding a second eye for the movie, but a whole list of graphic techniques is used to improve the S3D aspects of the movie. Kao (2009) asserted that a workflow for stereoscopic production must be discussed to build an entire pipeline which embedded depth consideration, then the team could author in stereo from the earliest stages of layout all the way through final quality control. An interactive approach pipeline is necessary as embedded depth control increased complexity of teamwork. According to Kao (2009), visual concept developing accord with depth script made by stereograhpers. The storyboard follows the depth script to illustrate the stereo illusion sketch. In the on-set section, depth control is the hub of graphic techniques tasks, digital artists create well-designed S3D works for subsequent stages to process correct depth perception. In the postproduction, left and right images are rendered, both images can then be edited to the final film (see Figure 2.21). Depth control drives more communication cycles mainly for graphic techniques departments such as modeling, texture/mapping, layout and 32

47 2.5 Colour Design in Filmmaking animation. The contributions of these departments directly determine viewer s experience, they must communicate with movie director and stereoscopy supervisor to improve the S3D aspect, making it more watchable and helping to avoid eyestrain. Moreover, they need to make sure that the S3D effect is integral to the telling of the story, not pulling the audience out of the immersive experience. Figure 2.21 Embed Stereoscopy Production Pipeline. Figure adapted from Kao (2009). 2.5 Colour Design in Filmmaking Film directors use colour to create contrasts between elements in the frame and communicate emotional ideas to the audience. A director works closely with the director of photography, production designer and costume designers to create a colour palette that fits the story of the film. Colour is a very important element to modern film and its application should be minutely thought out in wardrobe, sets, and lighting. Colour subconsciously communicates a host of messages to the audience and each of these messages can be controlled by the filmmaker through purposeful colour application. For instance, colour can be used to distinguish good 33

48 2.5 Colour Design in Filmmaking from bad, safety and danger, camaraderie and opposition, and far more. Intentional colour application to costumes, lights, and sets will go far in making a film stand out as excellent as opposed to amateurish. Night Shyamalan (2000) used red systematically in his film The Six Sense to indicate some connection with the world of dead. He intentionally inserted bright red objects as a symbol that something ominous was about to happen. (see Figure 2.22). 34

49 2.5 Colour Design in Filmmaking Figure 2.22 Colour design in the film The Sixth Sense. Images adapted from The Sixth Sense directed by Night Shyamalan (2000). Generally speaking reds are associated with strength, action, and passion. Yellows are cheerful, friendly, and sunshiny. Oranges are in between reds and yellows and most often communicate a cheerful warmth and hospitality that are bolder than yellows but more light-hearted than reds. Greens are affiliated with nature, peace, and refreshment though they can also communicate envy, sickness, or even mental instability. Blues are tranquil, relaxing, and calm though they can grow to communicate coldness or depression. Wachowski (1999) adapted green tint in the film The Matrix to 35

50 2.5 Colour Design in Filmmaking represent the opposing realities. The colour palette is based on the green text code that is by now a part of pop culture easily recognised. The film has another reality, the real world, which is represented by the use of a muted, blue tint. These colour choices are carefully selected to thematically and symbolically represent the different, opposing worlds. (see Figure 2.23). Figure 2.23 Colour design in the film The Matrix. Images adapted from The Matrix directed by Wachowski (1999). Balanced colours in foreground and background in the scene enhance the story and the mood of the film. Scott (2012) used colour thematically to present distant future world in Prometheus. In Figure 2.24, the characters uniforms are visually similar even if not exactly the same, navy blue, olive and greys against silver, greenish tint. The director not only dressed the characters thematically but also balanced the colours against the background of the sleek design of the ship. Figure 2.24 is another example 36

51 2.6 Colour Perception to show how the foreground colour can be visually homogenous with the background colour. Figure 2.24 Colour design in the film Prometheus. Images adapted from Prometheus directed by Scott (2012). 2.6 Colour Perception Colour appearance is generally considered together with the parameters of hue, saturation and brightness of the visual stimuli that are displayed in observer s field of view. Boff et al. (1986) indicated that the coloured visual stimulus observed by the viewer is specified by physical details of spatial properties such as size, shape, and location in the visual field, and temporal properties such as steady state, moving and pulsing, and their radiant power distributions such as spectral power distribution. The colour appearance of the visual stimulus derives from the experience that the observer gets and the judgment of colour appearance is directly influenced by the conditions and the environment that the visual stimuli are presented in. Nevertheless, 37

52 2.6 Colour Perception expressing the perceived stimulus is complicated, therefore, there are some models designed to describe the colour appearance precisely and universally. According to the CIE s (Commission International on Illumination) definition in 1932, colour can be defined as an attribute of visual perception consisting of any combination of chromatic or achromatic content which can be named by chromatic colours as yellow, orange, green or achromatic colours as white, gray, black, and it can be defined by the adjectives of dim, light, dark. Besides, perceived colour depends on the spectral composition of the radiant energy concerned by the observer according to the size, shape and surrounding of the stimulus areas. These stimulus areas can be both compared to each other as related colours or can be just judged as unrelated colours in which the circumstances will differ, and so their appearance will also differ from each other Hue, Saturation and Brightness Colour can be quantified by colour models. Having a specific set of values to describe the light itself is much more accurate than an ambiguous word like red. Rodney (2005) mentioned that colour space and colour model are used often in both colour management and computer imaging. A colour model or a colour space is the method of grouping numeric values by a set of primaries. Most colour models have three primary components such as RGB (Red, Green, Blue), HSV (Hue, Saturation, Value) and some colour models use more components such as CMYK (Cyan, Magenta, Yellow, Black). Joblove and Greenberg (1978) discussed the use of special colour spaces in computer graphics. In the RGB model, the primary colours, red, green and blue, can be added to produce the secondary colours of light, which are magenta, cyan and yellow. The importance of the RGB model is that it relates very closely to the way that human eye perceives colour, and it is a basic colour model for computer graphics because colour displays use red, green and blue to create the desired colour. The CMYK colour model is a 38

53 2.6 Colour Perception subset of the RGB model and is primarily used in colour print production. The K in CMYK stands for key as in four-colour printing cyan, magenta and yellow printing plates are carefully keyed or aligned with the key of the black key plate. HSV colour models were developed to be more intuitive in manipulating with colour and were designed to approach the way humans perceive and interpret colour. Since then HSV model is more intuitive than RGB model, it has consequently become ubiquitous throughout image editing and graphics software. This study will discuss colour based on a visual perception aspect to explore it s influence in stereoscopic depth perception, thus HSV model is the proper framework to consider colour design. Hue is used as an attribute of visual sensation to characterise the name of colours as red, yellow, green and blue which cannot be described other than its own. These four colours are the unique hues that are used in combinations to name other colour stimuli such as orange. Padgham and Saunders (1975) mentioned that hue is also understood as the variation in colour when the wavelength is changed. Secondly, brightness is the aspect of visual sensation that determines the level of emitted light from an area. Michel (1996) defined brightness as the variations in perception with the change in intensity. These variation ranges in brightness can be from very bright to very dim. Although in some cases lightness is used as brightness, the specific definition by Fairchild (2005) is that the brightness of an area judged relative to the brightness of a similarly illuminated area that appears to be white or highly transmitted. Additionally, the terms saturation and chroma are the most contentious in the literature of colour appearance. Chroma is one of the factors in colour appearance that involves a judgment between a chromatic colour and an achromatic colour of the same lightness. It is the colourfulness of an area judged as a proportion of brightness of a similarly illuminated area that appears white or highly saturated. It can be called the relative colourfulness; it approximately stays constant across the changes of luminance levels on the surfaces or objects which have the same hue. Saturation can be defined 39

54 2.7 Colour in Depth Perception as relative colourfulness as well, however, it is relative to its own brightness, whereas chroma is relative to the brightness of a similarly illuminated area that appears white. 2.7 Colour in Depth Perception Dresp and Guibal (2004) indicated that the colour of objects has a considerable effect on depth perception in the visual field. However, it is one of the most commonly debated cues, about which there are critical works. This section covers these works which concern the cues of depth perception and colour attributes and which affect the perception of depth. As Sundet (1978) mentioned, the effect of colour on objects apparent distances has long been known. This situation has mostly been practiced with highly saturated colours and with objects lying near each other. One of the earliest and well-known studies on this subject was conducted by Luckiesh (1918). The experiment applied the letters of red X and blue E and asked the subjects to move the letters back and forth in an apparatus until they came on to the same plane. He used tungsten lamp to illuminate the letters and the boxes were equipped with blue and red filters in order to obtain the coloured view. In the study, as the advancing quality of the red, most of the time red X was moved by the participants further away in order to make it appear on the same plane with blue E. This phenomenon has been explained by Sundet (1978) with an optical case under monocular and binocular depth perception. About the monocular theory, he stated that short-wave light refracts in the eye s optical media more than long-wave light, and because of this phenomenon, the equidistant sources of different colours cannot be simultaneously focused on the retina, which is called chromatic aberration. According to chromatic aberration, a short-wave light source is focused in front of a long-wave light source (see Figure 2.25). All theories of binocular colour-distance have chromatic aberration in common and this is taken the point of departure. From the 40

55 2.7 Colour in Depth Perception chromatic aberration phenomenon, it is well understood that wavelength is a stimulus that affects its perceived depth. Figure 2.25 Chromatic Aberration. Figure adapted from Sundet (1978). Before Sundet (1978) gave the information that colours in themselves have the quality of depth because of the refraction in the eye, Edwards (1955) mentioned in the conclusion of his experiment that there is no significant evidence of colour itself having the quality of depth. However, training and association may lead to seeing of some colours as near or far and may provide effective use of colour in art for the expression of depth. In opposition to Edwards (1955), Egusa (1983) explained an effect of hue on perceived depth, as green and blue difference is smaller than the red and green one, in which the red one appears nearer. Furthermore, in the first outdoor study, Mount et al. (1956) also mentioned hue effects on distance perception. In the experiment, they compared equal brightness coloured and gray papers under sunlight. According to the results, each colour is judged nearer than a gray that has equal brightness, and the hues appeared closer when viewed against the dark rather than in light. Additionally, Dengler and Nitschke (1993) pointed that when four colours (red, yellow, green and blue) are seen in front of a black background, long-wavelength colours appear in the front; however, in front 41

56 2.7 Colour in Depth Perception of a white background, the depth perception of colours are reversed. In that sense, the brightness of the background of a stimulus affects depth perception of the colours as well. However, there is still a gap on the depth perception of different hue combinations in terms of the stimuli and the background. Besides that, Mount et al. (1956) also pointed out that saturation and brightness contrast are more effective on depth perception than the difference in distance perception of one hue over the other. As another parameter of colour, saturation is also mentioned in many studies. According to Mount et al. (1956), the apparent position of a colour is advanced when the saturation of its colour is increased in contrast to its background. In this manner, it can be understood that the saturation difference between the stimulus and the surrounding also bears importance for depth perception. Besides, Egusa (1983) noted that if the higher saturated colour is red or green, they are judged nearer, but such an effect cannot be seen with blue colour. In a more recent study conducted by Bailey et al. (2006) with differently coloured three-dimensional objects of teapots and backgrounds on the computer screen, it is pointed out that similar results with high saturated warm or cool colours are obtained with less saturated colours in apparent distance perception. Camgöz, Yener and Güvenç (2004) named saturation as the secondary attribute of colour in judgment of nearness, whereas the brightness is the most dominant attribute. The apparent brightness and brightness contrast are also one of the most commonly studied cues in depth perception. In Michel s (1996) book, the aspect of perceiving brightness is said to be gamma movement. Gamma movement is when the object's brightness is increased and the object appears to advance toward the viewer from its initial fixation point. It returns to its former position when the brightness is decreased. The relation of brightness with coloured object distance perception was explained by Payne (1964). He described the insistency of a colour as the power of a colour to catch the eye and hold it steadily, and that insistent 42

57 2.7 Colour in Depth Perception colour appear nearer. The perceived relative distance may be affected by the relative brightness of the coloured objects. In another study considering the brightness effect on depth perception, Taylor and Sumner (1945) used fluorescent lamps and coloured papers (red, yellow, green, blue, white, black and neutral gray) at different brightness levels. Although the papers are in different hues, the experiment focused on the brightness differences. The subjects observed that the brightly coloured papers are drawn farther in the apparatus in order to equalise their apparent distances to each other. Thus, they stated that at a constant distance light colours appear nearer than dark colours. Furthermore, in another study by Johns and Sumner (1948), the same result of bright colours appearing nearer than dark colours at the constant distance is obtained. According to the results, the order of the colours from the one that appears nearest to the one that appears farthest is red, white, yellow, green, blue and black. The brightness effect also appears to be one of the contrast differences in adjacent colours referring to stimuli-surrounding relation. Payne (1964) mentioned that if a colour differs from the background more, it stands farther away from the background. Thus, one of the two colours has more similarity to the background will appear more distant than the other. In a similar way, Ichihara et al. (2007) claimed that contrast is an important cue for perceiving the depth of an object. It creates some illusions on the appearance of coloured surfaces in the manner of depth perception. They categorised contrast into area contrast and texture contrast. Area contrast is the difference between the average luminance of the surface area of an object and the average luminance of the background. When area contrast is low, the object looks far from the observer; similarly, it looks near when the contrast is high. Grandis (1986) defines simultaneous or reciprocal contrast as two areas of high contrast in adjacent positions altering the appearance of both. A light area next to a dark area appears lighter than it really is. Thus, this simultaneous contrast has the effect of darkening the dark colour and lightening the light colour more. The effect of two colours on each other makes lighter one to be perceived nearer than they are as an effect of colour 43

58 2.8 Colour Contrast kinetics. Colour kinetics is a property which makes colour to appear in the front or back rather than at its real location. This effect resulted from the degree of luminosity of each colour under lighting. The issue on brightness, darkness and contrast effects of colours has also been discussed under the subject of spaciousness and room size of the space. Taylor and Sumner (1945) stated that rooms done in white, light yellow, light green would appear smaller than they really are. However, Mahnke (1996) claimed that light or pale colours recede and increase the apparent room size whereas dark or saturated colours decrease the apparent size of the room. Clearwater (1986) mentioned that depth perception studies that conducted with apparatus are also done by fullscaled room with movable walls. A lightly coloured space appears larger, and there is a recession of blue that is highly dependent on its saturation. Michel (1996) also mentioned that regardless of its physical size, a bright room is perceived as more spacious. 2.8 Colour Contrast Colour contrast can be created between two or more colours in different ways. When two or more colours together display distinct variation, they are in contrast. Each colour can be adjusted to enhance its dissimilarity with the others, or surrounding colours may be selected to enhance conspicuous discrepancies between central colours. Contrast effects tend to enhance the individual colours by strengthening their differences, motivating a response in the individual who experiences them. Not all contrasts are created equal. Some are highly dramatic, while others are much more subtle, discovered only after considerable observation. To distinguish contrast effects from one another, Kopacz (2003) separated colour contrast into three groups: dynamic contrasts that are readily recognised by most people, such as contrast of value, contrast of hue and contrast of extent; subtle contrast reserved for more sophisticated situations, 44

59 2.8 Colour Contrast such as contrast of temperature, contrast of complement and contrast of saturation; complex contrast which offer the greatest creative challenges, such as contrast of simultaneity, contrast of succession and resolving contrast Contrast of Hue Hue contrast involves a combination of colours in which the distinctions of hue character are most pronounced. At least two colours but more often three or more different hues are used in combination. The effect is increased as colours are positioned farther apart on the colour circle, as long as they are not complementary. See Figure 2.26 for the colour wheel about contrast of hue. This effect of hue contrast is also stronger when more heavily saturated colours are used. Physical proportions of the hues creating a contrast do not need to be equal to be effective. Neither does their intensity of chroma. Some combinations in hue contrast are arranged so that one hue is given a larger percentage of area in the composition. While other colours serve a secondary role, that of enhancing the key colour. Other compositions create greater tension by maintaining a balance of all the colours in the composition. Another way to increase the effect of this contrast is to separate the colours with either black or white. White has the effect of making the colours of hue appear richer, while black will give them a greater luminosity. Either can increase the effectiveness of the hue contrast depending on the selection of colours in use. 45

60 2.8 Colour Contrast Figure 2.26 Colour wheel of contrast of hue. Figure adapted from Arditi (2005), 'Effective Colour Contrast' [Online]. Available: [1 January 2013] Contrast of Saturation Contrast of saturation encompasses the use of both intense and diluted colours in association with each other, usually working within one hue. The contrast is created only in subtractive colour, because it depends on the ability to manipulate saturation quantitatively. The spectral hues found in white light are uniformly of maximum saturation. This means the only way to reduce the intensity of an additive colour is to reduce the amount of light allocated to it. See Figure 2.27 for the example of saturation contrast. There are a number of ways to make a subtractive colour less saturated. One is to dilute it with white. This causes it to soften, as in pastel colour. Colour can also be diluted with black, creating rich jewel tones. Mixing with either black or white will also affect the value of the colour. Therefore, to maintain consistent value, a mixture of gray in equal value to the hue can be used to create a colour that is decidedly less intense. 46

61 2.8 Colour Contrast Figure 2.27 Example of saturation contrast. Figure adapted from Aaberg (2015) 'Colour Contrast - all about the difference' [Online], Available: February 2013] Contrast of Value The most forceful colour contrast is that of value, also referred to brightness or light-dark contrast. Contrast of value involves the interaction of light and dark colours in combination with each other. The effect can be done in shades of neutral colour, such as gray, black, and white, with some limited hue, or with a full range of hue. See Figure 2.28 for the example of contrast of value. To make the contrast stronger, the edges of the lightest and darkest values in the composition share common boundaries. The phenomenon of value contrast results from different levels of light reflected or absorbed by the image surface. Its greatest extreme is in black and white combinations, which a very dynamic compositions can be formed using a limited number of colours. 47

62 2.9 Colour Contrast and Depth Perception Figure 2.28 Example of value contrast. Figure adapted from JER (2012) 'The Importance of Contrast in Painting' [Online]. Available: January 2013] 2.9 Colour Contrast and Depth Perception O Shea et al. (1993) summarised two ways that contrast can affect depth perception. One is that low contrast increases uncrossed disparity to stimuli, resulting the objects appears farther than high contrast stimuli in stereoscopic depth. The other is that when depth is assessed as pictorial, low contrast stimuli appear far. In Fry and Bridgman s (1942) experiment, they used a stereoscopic technique and found that low contrast stimuli looked farther than high contrast stimuli. They concluded that contrast has unspecified effect on the stereoscopic mechanism. Contrast as a pictorial cue was discussed in the previous section. This section reviews several literatures that examined correlation between contrast and depth perception. Mount et al. (1956) concluded that high contrast stimuli appeared closer than low contrast stimuli. Farnè (1977a) also concluded that stimulus have lower contrast with the background appeared farther. The experiment required subjects to viewed a black and white target by monocular viewing at different distances and choose which was nearer. Farne s experiment allowed real differences in depth between the stimuli and the background, and included other depth cues to operate. 48

63 2.10 Depth Enhancement in Stereoscopic 3D Films The work that applied flat and pictorial stimuli is the study by Smith (1960). He produces contrast differences by a particular Munsell paper viewed against a background made from a different Munsell paper, to measure the depth of the patch. He then concluded that low contrast stimuli looked farther than high contrast stimuli and indicated that contrast can act as a pictorial depth cue. Egusa (1983) found that the less the depth, the lower the contrast between adjacent areas in a similar flat stimuli experiment. Troscianko et al. (1991) indicated that saturation of object stimuli decreases with distance by optical scattering and concluded that saturation and colour analogue of luminance contrast could act as depth cue. They found that changes in saturation were effective in producing apparent slant. Dresp et al. (2002) have shown that luminance contrast affects the depth perception of achromatic stimuli. Schor and Howarth (1986) also studied the effect of contrast on depth perception by using a stereo technique. They found that especially for low spatial frequency stimuli, low contrast stimuli appeared farther than high contrast stimuli. Viewed the stimuli monocularly, they found the same difference but reduced compared to binocular viewing. Rohaly and Wilson (1993) concluded that monocular and stereoscopic views are independent and additive and contrast has its effect on depth perception prior to the stereopsis mechanism Depth Enhancement in Stereoscopic 3D Films As discussed in Chapter 1, Digdia s (2011) report showed that S3D films have grown popularity in recent years and nearly 200 S3D titles have been released from 2008 to 2013 by Hollywood studios. This section reviews several famous S3D films (originally shot in 3D) and analyses the enhancement of depth effect by monocular depth cues in these films. 49

64 2.10 Depth Enhancement in Stereoscopic 3D Films Since Monsters vs. Aliens directed by Vernon and Letterman (2009), all feature films released by DreamWorks Animation are produced in S3D format. Purcell (2009) indicated that DreamWorks utilised Intel InTru3D 17 technology that enables animators to author the film in 3D for a more realistic 3D experience. Figure 2.29 shows a scene that creates perspective depth field by a moving Ping-Pong ball. The colour of the ball was designed in high-saturated warm colour and high brightness to create colour contrast against dim background. Figure 2.29 Depth enhancement in film Monsters vs. Aliens. Images adapted from Monsters vs. Aliens directed by Vernon and Letterman (2009). Friday the 13 th Part III directed by Miner (1982) was the first Paramount Pictures film produced in 3D since The film was shot by Arrivision Over and Under 3D camera that was used in Jaw 3D. An example of depth enhancement is that a character is killed and his eyeball pops out sending to the audience. The director applied Shallow Depth of Field that blurred the background in the scene to emphasis the pop-out object. (see Figure 2.30) 17 A brand developed by Intel to identify the content that may be viewed in S3D. 50

65 2.10 Depth Enhancement in Stereoscopic 3D Films Figure 2.30 Depth enhancement in film Friday the 13th Part III. Images adapted from Friday the 13th Part III directed by Miner (1982). (1982) Coraline directed by Selick (2009) is a stop motion animated mated film released in 3D. It was nominated as Best Animated 3D Movie of 2009 by People s Choice Award. According to Selick (2009), tthe he film was staged in a 140,000 square foot warehouse and animators shoot each frame from two slightly apart positions for its stereoscopy. In a snowy scene, the white background b sets off the main object (Coraline) to create clean contrast. The coat on the character was designed in a yellow warm colour to make the object appearing standout. (see Figure 2.31) Figure 2.31 Depth enhancement in film Coraline.. Images adapted from Coraline directed by Selick (2009). 51

66 2.10 Depth Enhancement in Stereoscopic 3D Films The Hobbit: An Unexpected Journey is a prequel of critical successful The Lord of the Ring trilogy directed by Jackson (2012). It was shot in 48 frame rate per second, which is twice than standard film frame rate, becoming the first feature film shot in High Frame Rate. The new technique allows cinematic visuals to better imitate what real world looks like and benefits 3D viewing experience smoother and less eyestrain. Figure 2.32 is a scene simply adapted size and position cues to create the depth of field with foreground, mid-ground and background objects. Figure 2.32 Depth enhancement in film The Hobbit: An Unexpected Journey. Images adapted from The Hobbit: An Unexpected Journey directed by Jackson (2012). Despite Cameron s (2009) strong opinion that 3D film requires higher frame rate to make strobing less noticeable, Fox studio still decided that Avatar would be done at 24 frame per second. The film was shot by 3D Fusion Camera System and Simucam technology that developed by Cameron (2009) to superimpose live action images and CGI images. The film broke the box office record in film history and became a milestone of 3D filmmaking. Figure 2.33 is a scene in Pandora world applied fancy colour with flying luminescent objects to create the visual immersion for audience. 52

67 2.10 Depth Enhancement in Stereoscopic 3D Films Figure 2.33 Depth enhancement in film Avatar. Images adapted from Avatar directed by Cameron (2009). Life of Pi directed by Lee (2012) won several film awards including Best Live-action 3D Film from 3D Creative Arts Award. Scenes in the film are well designed for 3D presentation, which seamlessly bring the audience in the middle of all actions. In a flying fish scene, the director adapted aspect ratio change to enhance stereo effect. The film ratio changed from 1.85:1 to 2.35:1 to allow fishes freak top and bottom of frame to enhance stereo presentation of the film even if not viewing it in stereo. (see Figure 2.34) Figure 2.34 Depth enhancement in film Life of Pi. Images adapted from Life of Pi directed by Lee (2012). 53

68 2.11 Conclusion 2.11 Conclusion The recent wave of S3D since Cameron s (2009) Avatar encouraged S3D content makers to consider stereoscopy from the outset of the production. It is therefore important to ensure the scenes in S3D content are designed properly for S3D to acquire proper depth perception. This chapter firstly reviewed human depth perception and the principle of stereoscopic vision to give the idea of depth creation, and discussed current S3D production pipeline and examples of colour design in films to outline the consideration of scene design and the use of colour in films. This chapter then reviewed literature in colour perception and related works of colour in depth perception. The preceding sections have discussed the existing findings of studies that have related to the work described in this thesis. The aim of this thesis is to discuss the influence of colour on S3D perception based on the findings that the colour of objects has a considerable effect on depth perception in the visual field (Dresp and Guibal, 2004). This thesis aims to evaluate the influences of hue, saturation and brightness respectively. Based on the conclusion that long-wavelength (warm) colours appear more standout than short-wavelength (cool) colours (Dengler and Nitschke, 1993), this work evaluates whether warm hues advantage the objects appear advance in computer-generated environment for S3D perception. The finding of saturation in depth perception about high saturated colours are obtained with less saturated colours in apparent distance perception (Bailey et al., 2006) indicates the attribute of saturation in judgment of nearness. This work is based on the finding to examine the saturation s attribute in nearness judgment in computer-generated environment by polarised projected stimuli. Camgöz, Yener and Güvenç (2004) named brightness is the most dominant attribute of depth judgment and Michel (1996) pointed about gamma movement that is when the object brightness increased and the object appears to advance toward the viewer. This thesis tests the point of view in 54

69 2.11 Conclusion CGI environment for S3D perception with reversed coloured stimuli to see if higher brightness object appeared standout. Evaluation methods applied in this thesis are reviewed and discussed in the next chapter. 55

70 3.1 Psychophysics CHAPTER 3 Psychophysical Methods, Image Quality Assessment and Stereoscopic 3D Presentation The aim of this thesis is to investigate human visual stereoscopic perception of different colour characteristics, thus to provide the evaluation of stereoscopic depth quality manipulated by different colour arrangements. The methodologies described in this thesis are psychophysical approaches to examine human visual perceptual thresholds and image quality assessment to evaluate the depth quality of stereoscopic images, and then the principles of stereoscopic polarised projection will be reviewed in this chapter. 3.1 Psychophysics The purpose of psychophysical experiments is to examine the relationship between the physical characteristics of a stimulus and the subjective human perception of it. A psychophysical experiment is based on responses of stimuli presented to the subject, who subsequently is required to make subjective response or judgment about the stimuli. The key idea of the relations between the physical and perceptual environment is based on Fechner s (1860) idea that body and mind are different reflections of the same reality. The processes of the brain are directly reflected in processes of the mind; Fechner anticipated establishing correlations between neuronal and perceptual events. The basic idea is to operate the physical stimuli systematically and acquire reports from observers about their perception of the stimuli. This section summarises the 56

71 3.1 Psychophysics introductory topics in standard reference works such as Guilford s (1954) book in psychometric methods, Baird and Noma s (1978) book about fundamentals of scaling and psychophysics, Nunnally and Bernstein s (1994) psychometric theory and Gescheider s (1997) book in psychophysics The Psychometric Function Wichmann and Hill (2001) indicated that the psychometric function relates the percentage of correct responses obtained by the subjects and is fundamental to psychophysics. To construct a psychometric function, a series of stimuli of different strengths or levels are presented to the observer. In this study, a two-alternative forced-choice task (2AFC) is performed for image stimuli. According to Fechner 's(1860) explanation, the observer in a 2AFC task is presented with standard and comparison stimuli side-by-side (could be presented with a temporal rather than spatial interval) in each trial. The subject must decide which of the two images is the more intense. A variety of comparison stimuli are presented in a random order, ranging those that are clearly noticeably different to the standard (where observers are expected to achieve a perfect score of 100% correct responses) to blank trials where no stimulus difference is presented (and where observers are expected to achieve chance performance of 50% correct responses). Chance performance is 50% because even when there is no difference between the standard and the comparison the observer will correctly guess which of the two stimuli is the comparison half of the time. The range of stimulus levels between those expected to produce chance and top scores is reflected by the slope of the psychometric function. The slope represents the number of correct respond as a function of the stimulus level of the comparison stimuli. The stimulus level at a score of 75% percent correct responses is usually adopted as the threshold value. The psychometric function slope provides a measure of how performance changes with changes in stimulus strength and is another parameter used to describe detection or discrimination performance. Changes in slope may 57

72 3.1 Psychophysics indicate fatigue or changes of attention in the observers, or the occurrence of perceptual learning effects. The psychometric function can also be analysed to reveal possible response bias in the observer Forced-Choice Procedure Developed by Fechner (1860), force-choice procedure requires observers to make judgements for correct responses within a limited set of alternatives. A standard and comparison stimulus presentation is repeated to ensure the observer is reminded of the stimulus variable they are required to discriminate, which reduces the likelihood of criterion shifts occurring. Observers are also required to choose a stimulus in trials where they are uncertain about which stimulus is the correct response. There are different psychophysical methods used to obtain estimates of threshold values, the following sections discuss their relative advantages and disadvantages Methods of Limits Blackwell (1952) described the psychophysical thresholds that the most basic function of any sensory system is to detect energy or changes of energy in the environment. This energy can consist of chemical, such as taste or smell, electromagnetic in vision, mechanical in audition, proprioception and touch or thermal stimulation. In order to be noticed, the stimulus has to contain a certain level of energy. This minimal or liminal amount of energy is the absolute threshold, which is the intensity that an observer can just barely detect. Another threshold, known as the difference threshold, is based on stimulus intensities above the absolute threshold. It refers to the minimum intensity by which a variable comparison stimulus must deviate from a constant standard stimulus to produce a noticeable perceptual difference. 58

73 3.1 Psychophysics In the method of limits, the intensity of stimulus is changed successively. Gescheider (1997) described the method that the stimulus initially is very weak, the observer should respond no difference, but gradually becomes noticeably different with the intensity increased in an ascending series; the stimulus is then changed from one that is clearly noticeable until it becomes unnoticeable in a descending series. The average of the intensity of the "noticeable difference" and the first "no difference" stimuli in the ascending trials, or vice versa in the descending trials, is recorded as an estimate of the absolute threshold. The difference between ascending and descending series described above is slight but systematic in thresholds. Therefore the results of the two series could be averaged to obtain the threshold measurements. In the experiment, stimuli are required to be presented either simultaneously or successively. The intensity of the comparison stimulus is changed in a series of steps while the intensity of the standard one is kept fixed. The comparison stimulus is either initially weaker or initially stronger than the standard. The series ends when the observer s response changes from weaker to stronger or vice versa. The difference threshold is then the intensity difference between the stimuli of the first trial on which the response differs from the previous one. Ascending and descending series are alternated and the results are averaged to obtain the threshold estimate Method of Constant Stimuli Laming and Laming (1992) described the method of constant stimuli that provides comprehensive psychometric measurements of changes in stimulus strength and the estimation of threshold value is determined based on a fully sampled function. In this method, the experimenter chooses a number of stimulus values, usually from five to nine, on the basis of previous exploration to encompass the threshold value. This fixed set of stimuli is presented multiple times in a quasi-random order that ensures each will 59

74 3.1 Psychophysics occur equally often. After each stimulus presentation, the observer reports whether or not the stimulus was detected or whether its intensity was stronger or weaker than that of a standard. Once each stimulus intensity has been presented multiple times, usually not less than 20 time, the proportion of "detected" and "not detected", or "stronger" and "weaker" responses is calculated for each stimulus level. Figure 3.1 Example of psychometric functional sigmoid curve The data are then plotted with stimulus intensity along the abscissa and percentage of perceived stimuli along the ordinate. The resulting graph represents the psychometric function. If there were a fixed threshold for detection, the psychometric function should show an abrupt transition from "not perceived difference" to "perceived difference". What experimenters usually obtain is a sigmoid curve (shown as Figure 3.1) that reflects lower stimulus intensities are detected occasionally and higher values more often, with intensities in the intermediate region being detected on some trials but not on others. This method is considered to generate the most precise and reliable data of the threshold measure. Simpson (1988) pointed out the method of constant stimuli is as efficient as adaptive approach that described in Section Each set of stimulus with different intensities expected to perform discriminations ranging from chance to a high correct response rate. The stimuli are presented at a random order and the observer is required to make a response after each trial. 60

75 3.1 Psychophysics In any case, the threshold thus occurs with a certain probability and its intensity value must be defined statistically. By convention, the absolute threshold measured with the method of constant stimuli is defined as the intensity value that elicits "perceived difference" responses on 50% of the trials. Two-parameter Weibull functions developed by Weibull (1951) are commonly applied for the experimental data in estimating psychophysical performance. The inverse of the function is computed to determine the threshold. The slope of the psychometric function provides a measure of the change in performance with changing stimulus intensity. Statistical techniques are normally used to estimate the parameters of the psychometric function. Although the method of constant stimuli is assumed to provide the most reliable threshold estimates, its major drawback is that it is rather timeconsuming and required a patient, attentive observer because of the many trials required The Method of Adjustment The simplest and quickest way to determine absolute and difference thresholds is to let a subject adjust the stimulus intensity until it is just noticed or until it becomes just unnoticeable or appears to be just noticeably different from, or to just match, some other standard stimulus. Fechner (1860) stated that the method of adjustment is distinct from the forced-choice procedure for the estimation of detection thresholds. This approach allows the observer to have the control of the stimulus variable, and to adjust the strength of the variable to find an optimal value. For example, the observer controls the intensity of a sound until it just becomes audible, and then the stimulus intensity is recorded to provide an estimate of the observer's threshold. Alternatively, the observer can adjust the sound from clearly audible to just barely inaudible, providing another estimate of the threshold. Typically, the two kinds of measurement are series in which 61

76 3.1 Psychophysics the signal strength is increased and series of decreasing signal strengths are alternated several times and the results are averaged to obtain the threshold estimate. The mean value is taken as the threshold from the repeated procedure. The main convenience of this method is that threshold values do not have to be obtained from the psychometric function. However, the highly subjective procedure and the observer s internal criterion may shift over the trial Adaptive Staircase The staircase method developed by Békésy (1960) is a modification of the Method of Limits. Adaptive testing procedure is used to keep the test stimuli close to the threshold by adapting the sequence of stimulus presentations according to the observer s response. Since a small range of stimuli need be presented, adaptive methods are relatively efficient. Simple up-down staircases reduce stimulus strength in subsequent trials when the response is correct, and increase stimulus strength when the response is incorrect. See Figure 3.2 for a curve example. The stimulus series starts with a descending set of stimuli. Each time the observer says "yes", the stimulus intensity is decreased by one step. This continues until the stimulus becomes too weak to be detected. At this point, the series reverses it's direction by increasing the stimulus intensity by one step. This procedure continues with increasing the intensity if the observer's response is "no" and decreasing the intensity if it is "yes". In this way, the stimulus intensity flips back and forth around the threshold value. Usually six to nine such reversals in intensity are taken to estimate the threshold, which is defined as the average of all the stimulus intensities at which the observer's responses changed. 62

77 3.1 Psychophysics Figure 3.2 Example of staircase curve When the subject cannot perceive the stimuluss change and report an incorrect answer, the stimulus intensity would increase. The intensity would decrease when the subject perceived the stimulus change correctly. This method concentrates the stimulus values, making it more efficient than the method of limits. However, this advantage refers the disadvantage that only a small number of target locations on the stimulus axis of the psychometric function can be estimated. Another problem with simple staircase procedure is that an observer may easily become aware of the scheme that runs the presentation. This could lead the observer to anticipate the approach of threshold and change the observer s response before the threshold is actually reached Choice of Psychophysical Methods Except method of adjustment, all methods of threshold determination do not allow the observe to control the stimulus intensity directly. As they rely on the experimenter's (rather than on the subject's) control, they provide a more standardised method of measurement. The method of constant stimuli is considered to provide the most reliable estimates but the number of trials required is also considerable. Another issue in constant stimuli is that, because only stimuli near the threshold provide relevant information, many of stimuli presented are too far away from the threshold to be of use. However, this inefficiency can be avoided by pre-testing in order to 63

78 3.1 Psychophysics determine the exact range of critical stimuli to use, tailored to the sensitivity of the observer. In the staircase method, most of the stimulus values are concentrated in the threshold region, making it a more efficient method than the method of limits and constant stimuli method. However, the major issue of this method is that the observer will aware the scheme that governs stimulus presentation and anticipate the approach of threshold and change the response before the threshold is actually reached. To overcome this problem, the interleaved staircase can be applied. The two staircases may also be interleaved in a random rather than regular sequence to prevent the observer from figuring out which staircase to expect from trial to trial (Cornsweet 1962). Considering the precision and reliability of the psychophysical parameter estimates, this thesis applied the method of constant stimuli as the psychophysical procedure to construct the psychometric functions. This method also provides the robustness of variability estimates, which is critically important for the psychophysical data collection. This thesis investigates the processing of different colour arrangements in stereoscopic perception in individual observers and the generation of reliable estimates allows the comparison of patterns of psychophysical performance across observers. Weibull functions were fitted to the data obtained in order to generate parameter and variability estimates. In spite of the reliability and precision of the psychometric function estimates obtained from constant stimulus methods, this thesis applied image quality assessment technique described in Section 3.2 to evaluate stereo image quality to generate data to support or criticise the results from psychophysical experiments. Thus, the comparison of psychophysics and image assessment maximized the referential value of the experiment results conferred by both approaches. The image quality assessment methods are described in the following section. 64

79 3.2 Image Quality Assessment 3.2 Image Quality Assessment Very often the quality of an image needs to be evaluated and examined. Image quality assessment quantifies the quality of an image normally by subjective test session or by objective computational metrices. Many metrices based on different principles have been developed to predict how much of the distortion will be observed by user. Subjective methods for digital image quality measures are defined in the International Telecommunication Union Radiocommunication Sector Recommendations (ITU-R Rec.). The recommendations advise the assessment procedure, environment, equipment and result analysis. The principle of subjective methods is that a group of observers judge the quality of an image or video being presented to them. Subjective methods are considered the most accurate method in determining how much distortion or difference of image can be perceived. In the ITU-R recommendation (ITU-R-500-7, 1997), experimental methods are described to assess the quality of impaired still images and image sequences. The following sections describe the three main different approaches: the double-stimulus-continuous-quality-scale method (DSCQS), single-stimulus method and stimulus-comparison method DSCQS Method Wolf and Pinson (2002) indicated that DSCQS method of performing subjective test is widely accepted as an accurate test method. In DSCQS method, observers assess a series of image pairs for the overall image quality. Each pair consists of a reference image (non-impaired) and a test image (impaired). Observers assess the overall picture quality separately and eventually the assessment results are differences of scores between the reference and test image. 65

80 3.2 Image Quality Assessment Sessions normally last up to half an hour. The assessor is presented with a series of images pairs in random order with required combinations of impairments. At the end of the sessions, the mean scores for each test condition and testt picture are calculated. A test session comprises a number of presentations. For a single observer, the observer is allowed to switch between A and B images until the observer has decided on the quality associated with each image in each presentation. Typically, The assessor may choose to do this two or three times for up to 10 times. See Figure 3.3 for the presentation structure of test material. Figure 3.3 Presentation structure in DSCQS method. Figure adapted from ITU-R (1997). The overall quality of each presentation is assessed through inserting a mark on a vertical scale by observers. The vertical scales are printed in pairs to accommodate the double presentation of each test picture. A continuous rating system is provided in the scale to avoid quantizing errors which is divided into five equal lengths that correspond to the normal ITU-R five-point 66

81 3.2 Image Quality Assessment quality scale. The five-point quality scale is associated with terms categorising the different. Figure 3.4 shows a section of a typical score sheet. Figure 3.4 A five-point scoring sheet of DSCQS method Single-stimulus Method In the single-stimulus method, the assessor provides an index of presentation contented with a single image or sequence of images. A series of assessment trials can be include in the test. All images are presented in random order and preferably in different random order for each observer. Categorical scales that assess image quality and image impairment have been used most often. See Table 3.1 for the ITU-R scales. Semantic terms in a set of categories are assigned with an image or a sequence by observers in adjectival categorical judgments. The terms may reflect observers judgment. Half grades sometimes may be used and some special cases may assess text legibility, reading effort and image usefulness. 67

82 3.2 Image Quality Assessment Table 3.1 Five-point scoring scale and associated terms in Single-stimulus method Stimulus-comparison Method In stimulus-comparison methods, two images, standard and comparison, are presented and the observer judges based on an index of the relation between the two images. The resulting images or image sequences are combined to form the pairs that are used in the assessment trials. Stimulus-comparison method assesses the relations among stimuli more fully when judgments compare all comparison stimuli with standard stimulus. However, if this requires too many observations, it is possibly divided or to use a sample of all possible pairs. Semantic terms are used in adjectival categorical judgement, observers assign the relation between a pair to one of a set of categories. These categories may report the perceptible differences such as same, different, the direction of perceptible differences such as less, more, or judgements of extent and direction. See Table 3.2 for ITU-R comparison scoring scale. 68

83 3.2 Image Quality Assessment Table 3.2 Scoring scale and associated terms in Stimulus-comparison method -3 Much worse -2 Worse -1 Slightly worse 0 The same +1 Slightly better +2 Better +3 Much better This method generates a distribution across scale categories for each condition pair. The responses are analysed depends on the judgment made, and the information required, such as just-noticeable differences and ranks of conditions Choice of Subjective Methods Single-stimulus method assesses the image quality in the stimulus set individually. In stimulus-comparison method, a series of images pairs is applied. These image pairs include standard and comparison images in the stimulus set. In this procedure, observers assign a relation between the two images for each image pair. The same single-stimulus and stimuluscomparison methods can be used to assess impairment. These scaling methods use different grading scales to assess perceptual image quality. In DSCQS, the scale is often associated with semantic terms such as excellent, good, fair, poor, and bad to guide the observer. For single-stimulus scaling and stimulus-comparison scaling, it is usually applied rating scales such as verbal or numerical categories. The subjects report the perceived picture quality, the relation, or the impairment between two images by placing the presented stimuli in one of these categories. This thesis applied the procedure of stimulus-comparison method with the scoring scale of DSCQS method to evaluate image quality with reference 69

84 3.2 Image Quality Assessment and test images presented side by side to observers. The subjects observe reference and test images at same time to avoid the flaw of keep the reference image in mind procedure Experimental Conditions Evaluation methods as described above are used to measure the inputoutput relationship between manipulated imagery and human visual sensations. The sensation is expressed as a response of image quality gradations using qualitative terms, such as excellent or bad image quality. Unlike in threshold experiments where the unit of the rating scale can be defined as just-noticeable difference, the image quality degradation scale as used in supra-threshold experiments is an ill-defined scale. The image quality judgments can be affected by contextual effects such as image content, presentation order and stimulus spacing (de Ridder, 2001; ITU-R- JWP10-11Q, 1998). Threshold experiments have mainly been conducted with simple stimuli such as sinusoidal grating patterns. These stimuli have been useful in perceptual studies, such as measuring display fidelity. However, image quality in terms of appreciation cannot be addressed with such simple stimuli. The trend in image quality studies is towards using complex natural scenes. The effect of a specific degree of impairment on image quality is not necessarily the same for images with different content. For example, it depends on the information that is lost or on how annoying the distortions are in a particular region of an image. 70

85 3.3 Statistical Analysis on Visual Experiment 3.3 Statistical Analysis on Visual Experiment T-Test in Psychophysical Experiments and Visual Grading Having obtained the data by conducting an experiment, the experimenter should treat the data statistically and analyse the results. Rajamanickam (2002) mentioned that even if the same stimulus is presented a hundred times and a hundred responses are recorded in psychophysical experiments, some might have the same value and the rest might have fluctuated up and down. Therefore, the mean values are worked to explained meaningfully and other statistical treatments are given to the data to present. The mean values can show the differences among groups, however, in psychophysical experiments, more often the difference between the groups is not so obvious, in these circumstances, use of a t-test can help to decide whether the difference between the groups is real or whether is due merely to chance fluctuations from one time of testing to another. The t- test enables the assessor to decide whether the mean of one condition is really different from the mean of another condition. There are two types of t-test, dependent-means t-test and independentmeans t-test. When the same subjects participate in both groups of the experiment, dependent-means t-test, also known as the matched pairs or repeated-measures t-test will be adopted. When one group of subjects participate one condition in the experiment, and the other different group of subjects participate the other condition, the independent-means t-test, also known as independent-measures t-test will be adopted. In a repeatedmeasures experiment, two conditions are tested by the same group of subjects. Each subject does both condition A and B and the assessor will expect each subject to behave fairly similarly on both occasions because most of the characteristics of each subject, such as age, education, gender, remain the same on both occasions. As long as the performance measure is reliable and reasonably stable from one time of testing to another, there should be a strong relationship between a subject's performance in condition 71

86 3.3 Statistical Analysis on Visual Experiment A and condition B. In a independent-measure t-test, there are two groups of different subjects: one group for the condition A and the other different group of subjects for the condition B. Even if the assessor treated both groups in exactly the same way, variation will be expected between performance in condition A and condition B as subjects in different ages, educations and genders remained constant across the two conditions. In a repeatedmeasures design, the effect of the manipulation has to show up against the variance of individuals' performance from one time to another, and the fluctuations which are relatively minor. With an independent-measures design, the effect of experimental manipulation has to show up against the inherent variance of the differences between individuals, which is greater than the differences between the same individuals at two different times. What this means in practice is that the experimental manipulation is more easily with a repeated-measures design. In essence, both types of t-test can be applied when there are two sample means and the difference is believed existed in between. For visual grading experiments, each image is graded in one or more respects by a number of observers who select a score reflecting the general quality of the image or the fulfilment of a specific criterion such as the visibility, colour or three-dimensional recognition. A variant of the method, intended to increase the sensitivity to small differences in image quality, involves simultaneous viewing of two images, where the score is meant to express a comparison of the two images, such as +2 for "certainly better in right image than in left image", +1 for "probably better in right image than in left image " 0 for "equivalent", -1 for "probably better in left image than in right image", -2 for "certainly better in left image than in right image". This judgement may refer to a general concept of image quality or to a single well-defined criterion. The primary data required for t-test is the mean scores of the two groups and the number of participants. The equations is: 72

87 3.3 Statistical Analysis on Visual Experiment Here X1 is the mean scores of the experimental group X2 is the mean score of the control group. SS1 is the sum of squares of the first group. SS2 is the sum of squares of the second group. N1 is the number of participants in the first group. N2 is the number of participants in the second group. The experimenter will see whether the mean score difference between the two groups is real or spurious. The 't' test is a criterion to judge this assumption. The 't' may be small or very small or large. Whether is small or large it may determine the level of difference between the means of the two groups. To test this, the assessor must make use of Table IV of Fisher that is a guidance of distribution of 't'. This table may be found in every statistic book under the heading of 'Table of t'. The Degrees of freedom (df) is found vertically (column) and the P horizontally (row). The assessor will have to find out the P value associated with the df and the t value from the table. If the P value is associated with less than 0.05, the t value calculated from the experiment is greater than the table value and there is significant with 0.05 level and the mean difference between the two groups is not spurious. In other words, if this experiment is conducted for 100 times there is only less than five chances of being no difference between the two groups in the mean scores occurs. If the P value is associated with more then 0.05, there is no significant difference between the two groups Univariate Analysis of Covariance Covariance is a measure of linear association between two variables. Analysis of covariance (ANCOVA) is a general linear model which blends 73

88 3.4 Stereoscopic 3D Projection ANOVA 18 and regression. ANCOVA evaluates whether population means of a dependent variable are equal across levels of a independent variable. The assessor applies Univariate General Linear Model (Univariate GLM) in SPSS 19 to implement ANOVA AND ANCOVA statistical procedures. There are several key assumptions that underlie the use of ANCOVA and affect interpretation of the results. The particular important one in evaluating the appropriateness of ANCOVA is the homogeneity of regression slopes. The standard linear regression assumes that the slope of the covariate is equal across all treatment groups. Assessors use ANCOVA when comparing two or more regression lines to each other. ANCOVA will tell whether the regression lines are different from each other in either slope or intercept. Valentini et al. (2014) indicated in a study about psychophysical and cortical responses to threatening stimuli that ANCOVA is the most powerful statistical approach for experiments in which subjects are assigned randomly to treatment groups. ANCOVA can be used to increase statistical power 20 by reducing the within-group error variance. The test used to evaluate differences between groups is F-test computed by dividing the explained variance between groups by the unexplained variance within the groups. If this value is larger than a critical value, the assessor concludes that there is a significant difference between groups. 3.4 Stereoscopic 3D Projection To simulate a 3D cinema viewing experience, this thesis performed a polarised projection viewing of stimuli, which is based on the principle of current 3D cinema standard. 18 Analysis of variance (ANOVA) is a collection of statistical models used to analyze the differences among group means and associated procedures. 19 SPSS is a software package used for statistical analysis developed by IBM. 20 The ability to find a significant difference between groups when on exists. 74

89 3.4 Stereoscopic 3D Projection Polarised Projection Woligroski (2010) pointed that dual-projector polarised displays are used in most 3D cinemas, such as RealD, IMAX and Dolby3D. This cinema formats applied a variation of this 3D display technology. Polarised 3D projection triggers the illusion of stereoscopic 3D images by limiting the light that comes through polarised filters. The low-cost eyeglasses that the viewer wears are a pair of orthogonal polarising filters. Each filter only passes the light that is similarly polarised and blocks the orthogonally polarised light. On the viewing screen, two images are projected by two projections filtered by polarising lens. The light reflected from the screen may lose a bit of polarisation, but this issue is eliminated by utilising a silver or aluminised screen. (see Figure 3.5) Figure 3.5 Polarised Projection Technique. Figure adapted from Woligroski (2010). 75

90 3.4 Stereoscopic 3D Projection To implement a low-cost system to display stereoscopic 3D images, the basic equipment includes a pair of aligned projectors, a silver screen, polarising filters, and a computer with a dual-head graphic card (or connection to a video splitter) Projectors In stereoscopic 3D projection, one projector displays the left-eye information, the other displays the right-eye information, both at standard refresh rates. A polarising filter mounted in the optical light path of each projector ensures that the correct information passes though its corresponding filter in the pair of passive stereo glasses. The two projectors may generate different refresh rates and alternate the visual information between left and right eye. The issue of using CRT projection is that CRT projectors produce refresh rate normally between 60Hz to 85Hz, which is not high enough for high quality stereo display. The main advantage of CRT projectors is they produce unprecedented visual fidelity with the highest resolution screen images. Moreover, the filter mounted in the projector can be moved away for normal non-stereoscopic projection. The main advantage of utilising Liquid Crystal Display (LCD) projectors for S3D projection is they are allowed to build internal polarisation to its corresponding left and right eye information and external filters are not required. The internal polarisation maintains 70% of the original light, making this technology the most efficient way to obtain high stereo lumen brightness levels. However, the vertical refresh rate produced from LCD projectors is not high enough to display a high-quality stereo projection with a single projector. The recommended refresh rate is at 96Hz above. In comparison, Digital Light Processing (DLP) projection technology bounces light off a digital micro mirror device instead of passing light through a liquid crystal material. The micro mirror device contents tiny mirrors that each reflects a single pixel in the resolution of the projected image. The main advantage of DLP technology for S3D projection is brighter and smoother 76

91 3.5 Scene Design for Stereoscopic 3D Presentation projection, which is beneficial when polarised filters eliminating brightness Linear and Circular Light Polarisations Different information can be directed to left and right eye to create stereoscopic depth perception by organising the polarisation oppositely. Only one orientation of light emerges when passing light through a polariser. Linear polarisation polarises light in a single direction, north to south or east to west. The viewer wearing linear-polarised glass may change the orientation by tilting his or her head, which result orientation of polarisation doesn t match the orientation of the polarisation filter mounted on the projector. The stereo information then is lost. However, linear polarisation nevertheless produces precise image separation between the left and right eye and is cost-effective for stereoscopic application. Circular polarisation breaks the limitation of viewers head tilting. The light is not polarised in a single direction, which allow viewers can change his or her viewing angle relative to the stereoscopic projection display. This high-end technology requires greater precision and a tighter match between the polarising filters in the light path of the projectors and those in the glasses. The current RealD cinemas utilise circular polarisation to present S3D films. 3.5 Scene Design for Stereoscopic 3D Presentation Foreground, Mid-ground and Background Objects The foreground, mid-ground, and background in a composition are generally divided into three planes. The foreground of a composition is the visual plane that appears closest to the viewer, while the background is the plane in a composition perceived furthest from the viewer. The middle ground is the visual plane located between both the foreground and background. The scale of these components often correlates to the dominance in an 77

92 3.5 Scene Design for Stereoscopic 3D Presentation image. The foreground is often the most dominant due to the larger perceived scale of the images objects. Figure 3.6 illustrates the concept of foreground, mid-ground and background stages in scene design. Visually, we often refer to the scale of one object to another. As objects come forward in space, towards the viewer, they appear larger. As they recede into the background, their scale gets perceptually smaller. Figure 3.6 Foreground, mid-ground and background areas in scene design Scene Design for Comfort Viewing For a comfort stereoscopic viewing experience, vergence and accommodation in stereopsis have to be considered. Sun and Holliman (2009) indicatedd viewer s Geometry of Perceived Depth (GPD) on stereoscopic display should be limited into a defined volume, the so-called Comfortable Viewing Range (CVR). In stereopsis, the vergence and accommodation generate stereo depth and the link between these points is thought to be not to exceed the range of CVR. Viewers may experience eyestrain or visual discomfort if the vergence point was a long way off the display plane. (see Figure 3.7) 78

93 3.5 Scene Design for Stereoscopic 3D Presentation Figure 3.7 Comfort stereoscopic viewing range discussed in Sun and Holliman s (2009) work which indicates that objects are encouraged to arranged in the comfort viewing range calculated. Figure is adapted from Holliman (2009). To guarantee a comfort stereoscopic viewing experience, all stimuli in this work were designed in Autodesk Maya 2011 by applying visualisation tools to bound the scene and ensure all objects located within CVR area Stereoscopy in 3D Application Packages The stereoscopy in 3D graphic design packages is to set up virtual stereo cameras to simulate the human perceptual image by taking left and right eye s perspectivee views and use that information to construct a 3D representation of the real world. Virtual cameras in computer-generated environment mimic the principle of real world cameras. They have focal length, focus distance, depth of field and f-stop, which have to be defined by adjusting parameters and values in the software. Every aspect can be defined and designed in the virtual world and this provides the benefit of ultimate creative control. Cameras can be 79

94 3.5 Scene Design for Stereoscopic 3D Presentation duplicated and placed anywhere without physical limitations. Some problems happened in real world cameras such as fringing, breathing and lens distortion are not presented in virtual cameras. Exploring the parameters available in real world cameras would be helpful to create prefer stereoscopic images by virtual cameras. In photography, foreshortening effect is useful for S3D as it provides more depth information for the human visual system to reconstruct the imaginary into 3D. The camera model in Autodesk Maya is a very comprehensive representation of a virtual camera. Maya uses 6 parameters to construct the viewing frustum of the camera. The camera frustum is composed by left, right, top, bottom, near and far parts and covers perspective viewing area of the camera (see Figure 3.8). These 6 values form a projection matrix that indicates a range of field for safe stereopsis when constructing the scene. Far Clip Plane defines the visible furthest point from the camera while Near Clip Plane defines the closest visible point from the camera. Other values in virtual camera are also critical to define stereo imagery. Film Offset parameters represent the shift of film in the back of the camera and specify a positive or negative value on the horizontal shift. Focal Length is either specified as a field of view or by focal length. Film Aperture refers to the format of the projected image area. Film Back simulates a real world camera with a virtual film onto which images are projected. 80

95 3.5 Scene Design for Stereoscopic 3D Presentation Figure 3.8 Six parameters that construct viewing camera frustum in Autodesk Maya. The stereoscopic camera model in Maya (since Maya 2009) consists of three cams in the scene: centre cam, left cam and right cam. Left cam is identical to the centre camera. Centre cam is an attribute pointing to the main camera but it is not influential for stereo adjustments. Two key parameters play a key part in the stereoscopic camera model: Zero Parallax defines the point of convergence from left and right cams. Interaxial separation defines the distance between left and right cams. Stereo artists specify the interaxial and zero parallax values to shift cameras to achieve the appropriate parallax. 81

96 3.5 Scene Design for Stereoscopic 3D Presentation Figure 3.9 Stereo camera attributes in Autodesk Maya Visualised tools in stereoscopic camera model help digital artists to measure the intensity of stereoscopic perception. Zero Parallax Plane represents as the movie screen in the scene to indicate which objects stay in front or behind the screen. The size and the location of the plane changed following the parameters of zero parallax and interaxial distance defined by digital artists. (see Figure 3.9) Stereoscopic Rendering Stereoscopic rendering can be performed simply with Maya that supports batch rendering from multiple cameras with little setup required. This thesis utilised Maya Software Renderer to produce all image stimuli. Rendering stereoscopic images is to trick the brain into thinking there is depth in a flat 2D image, by using two cameras that are offset just as our eyes are, then taking the two renders and merging them together. By using two cameras, we can take the separate renders and merge them in a 82

97 3.6 conclusion compositing package, to create a stereoscopic sequence. Generally, we need a centre camera to align and direct shots, and to possibly render out for traditional 2D viewing. Each camera has slightly offset view (see Figure 3.10). Figure 3.10 Images rendered from stereo camera rig in Autodesk Maya. From left to right, the images were captured from left camera, centre camera and right camera HSV in Maya HSV (Hue, Saturation and Value) is one of the two colour modes in Maya. Hue refers to a value 0 to 360 or a colour found on a standard colour wheel. Saturation is the amount of gray mixed into the colour, measured from 0 to 1 in Maya. The saturation value determines how pure or how diluted the colour will be. Value determines how bright or how dark the colour will be measured from 0 to 1, which is the overall brightness. HSV mode allows a colour s intensity and brightness to be adjusted independently and this is a effective way to achieve a desired colour. 3.6 Conclusion This chapter reviewed experimental methods utilised in this thesis and the construction of polarised projection applied in the experiments. The psychophysical procedures and depth quality assessment used in this thesis are outlined as following. Any exceptions are stated in the methods section of each chapter. 83

98 3.6 conclusion In Chapter 4, full psychometric functions were generated using a 2AFC constant-stimulus psychophysical procedure, based on seven stimulus levels and 50 trials per level. Estimations of the 50% PSE and 75% threshold value and it s 95% confidence interval were generated using the approach described in Winchmann and Hill's (2001) work. Observers were presented with a pair of side-by-side stereoscopic images (reference and test) and were required to wear a 3D glasses to discriminate which image was perceived more distance between foreground object and background object. The order of presentation was randomised across trials. In Chapter 5, stereoscopic depth quality was measured by stimuluscomparison method based on ITU-R scales. The observers were presented side-by-side with a pair of reference and test images and were required to wear a 3D glasses to judge the depth quality on the test image with a fivepoint quality scale associated with terms bad, poor, fair, good and excellent. The polarised projection was run by two same model DLP projectors with 1024*768 native resolution, 2600 ANSI Lumens and 2500:1 contrast ratio to project left and right images on the same screen synchronously. Polarised filters were set in front of projectors to polarise projection lights. A 100-inch silver screen was then used to display stimulus images by the projection. The actual projected image size on the screen was 1460mm x 1380mm. Viewers were asked to sit exactly 300 cm in front of the silver screen to warrant a comfort viewing experience. The experimentations based on the methods and principles described in this chapter are presented in Chapter 4 and Chapter 5. 84

99 4.1 Introduction CHAPTER 4 Psychophysical Evaluation of Saturation, Hue and Brightness on Stereoscopic 3D Perception 4.1 Introduction This chapter explores saturation, hue and brightness on stereoscopic 3D perception from a psychophysical viewpoint. It considers the way in which S3D perception is influenced by different decisions of colour between foreground and background, and identifies the thresholds of S3D perception in colour contrast adjustments. The experiments are carried out, firstly, to examine whether changes of saturation, hue and brightness influence observers psychometric functions of stereoscopic depth perception in standard trial and reversed trial. The colour arrangements in standard trials are based on previous studies supporting the idea that high-saturated objects appear standout than low saturated objects (Egusa, 1983; Bailey et al., 2006; Camgöz, Yener and Güvenç, 2004). Long-wavelength colours appear closer to the observer than short wavelength colours (Sundet, 1978; Egusa, 1983; Dengler and Nitschke, 1993). High brightness objects appear closer than low brightness objects (Payne, 1964; Michel, 1996). The reversed trials are then swapped the colour values in foreground and background in the standard trial to testify and compare. These previous studies examined depth perception under the monocular or binocular viewing conditions but did not extend the results to the effect when viewing under stereoscopic 3D projection. A key advance of the work in this experiment is that the effect of saturation, hue, brightness on the strength of a stereoscopic 3D perception is explored and contrast between foreground and background was controlled to acquire comparable 85

100 4.2 Hypotheses data. For instance, in the saturation section, the standard trial carried out the effect of increasing saturation in the foreground was examined whereas in the reverse trial the effect of decreasing the saturation in the background was tested. This work subsequently compares the psychophysical thresholds between standard and reversed tests. This work aims to apply the findings of previous studies (e.g. Dresp and Guibal, 2004) suggesting that depth perception is affected when alterations are made in saturation, hue and brightness, to a stereoscopic 3D viewing condition when a current 3D cinema projection system is applied. Six trials run in this experiment to examine several conditions. (see Table 4.1) Table 4.1 List of trials in the experiment Section Trial Colour arrangement Saturation increased in foreground and Standard background colour remained unchanged. Saturation Foreground colour remained unchanged and Reversed saturation decreased in background object. Warm hue in foreground with cool hue in Standard background. Hue Cool hue in foreground with warm hue in Reversed background. Higher brightness in foreground with lower Standard brightness in background. Brightness Lower brightness in foreground with higher Reversed brightness in background. 4.2 Hypotheses Based on the previous studies described in Sections 2.7 and 4.1, the results in these experiments have been designed to answer the following hypotheses: 86

101 4.3 Methods 1. High saturated foreground scenes perform higher discrimination in stereoscopic 3D depth perception than low saturated foreground scenes. 2. Warm (long-wavelength) coloured foreground scenes perform higher discrimination in stereoscopic 3D depth perception than cool (shortwavelength) coloured foreground scenes. 3. High brightness foreground scenes perform higher discrimination in stereoscopic 3D depth perception than low brightness foreground scenes. 4. Colour contrast is more dominant in stereoscopic 3D depth perception than saturation, hue and brightness. 4.3 Methods This study employed psychophysical experiments for the measurement of threshold values. The method of constant stimuli was chosen as described in that Farell and Pelli (1998) indicated the method is considered to provide robust and precise estimations of threshold and other parameters. Psychometric functions were constructed based on seven stimulus levels with 50 trials per level. 2AFC procedure was employed to minimise the likelihood of criterion shifts and also reduces the possible effects of intervalselection bias (Johnson et al., 1984). See more description of psychophysical methods in Section Observers Seven observers, three male and four female, aged 25 to 34 with a mean of 29 years, took part in this experiment. Subjects were not aware of the purpose of the experiment and they were all non-expert; their normal work is not concerned with stereoscopic perception. All participants met the 87

102 4.3 Methods minimum criteria of stereoacuity at 40 sec-arc and passed the Ishihara Colour Blindness Test. Participants were required to read an instruction paper before the trials. The judgement request for each observation stated in the instruction Select the image you perceive more stereoscopic depth distance between foreground and background objects. The instructions contained example images to illustrate the definitions of the foreground and background objects and stereoscopic depth distance. Each participant finished 350 observations in each section and received a gift as the reward. All participants were postgraduates recruited from within the university and varied in nationalities and culture backgrounds. Two Taiwanese, two Korean, one Syrian, one Egyptian and one Chinese had participated the experiment. Some studies indicated that cultures may influence colour interpretation (e.g. Bratu 2010). For example, red is the colour of love in most cultures. Chinese brides wear red for their wedding, and red roses are the most common gift for Valentine's Day. Red can also be the colour of communism. The flags of China, Vietnam and the former Soviet Union are red. In Christianity, green and red are associated with Christmas and Satan is also most of the time represented by the colour of red in icons and popular culture. Masuda (2009) suggested that people developmentally acquire perceptions of a three-dimensional world in accordance with their experiences with the surrounding environment. Once people acquire this specific perceptual pattern in the three-dimensional world, they apply the same rules even when they observe the visual representation in the twodimensional field. However, observers in this experiment were asked to compare the specific stereoscopic depth distance between the foreground and background objects instead of to interpret the whole depth of field in the scene, and there is no studies indicating the evidence of relation between culture and the depth perception that manipulated by colours. Therefore, observers culture background was not involved as a intervening factor in this experiment. 88

103 4.3 Methods Stimuli Gescheider (1997) indicated that constant stimuli method is the procedure of repeatedly using the same set of stimuli, usually between five to nine different values in the set throughout the experiment. This study applied seven different values of stimuli. The lower end which is the stimulus that more difficult to detect the difference, and the upper end which is the stimulus that can always be detected the difference. All stimuli designed and rendered in Autodesk Maya The general stimulus images design method is described in Chapter 3, Section 3.5. Specifically, each stimulus image is a combination of 2 components, foreground object and background object. Both objects are poly planes in square shape with a Lambert material Scene design To guarantee a comfort stereoscopic viewing, the scenes were designed with considerations of image size, projection size to calculate maximum parallax to avoid divergent viewing of left and right images. When the parallax was within the value we calculated, the comfort stereoscopic viewing would be guaranteed. The foreground square plane with the dimensions 4.9cmx4.4cm (centimetre is the default unit in Maya) has been placed parallel with the background square plane with the dimensions 43.2cm 35.9cm. With the visualised display screen in the scene, the foreground plane has the distance 6.88cm in front of the screen; the background plane has the distance 7.27cm behind the screen. The whole scene is located within stereoscopic safe volume that suggested by the visualised volume in Autodesk Maya. The scene was then captured by the virtual stereo camera rig and rendered into left and right images by Maya Software Renderer. (see Figure 4.1). 89

104 4.3 Methods Figure 4.1 Stimulus scene design for a comfortable stereoscopic viewing. The red plane in the scene represents display screen and is invisible in the final rendered images Stimulus Images and Colour Values Seven levels of stimulus image tested in each section are shown in Appendix A. They all have the same frame composition with the image size 445x334 pixels. In each trial, the observers see a standard stimulus and a comparative stimulus presented side by side on a 1024x768 pixels canvas. (see Figure 4.2) Figure 4.2 Dimensions of foreground colour palettes, background colour palettes and canvas. 90

105 4.3 Methods Colour values were defined in HSV colour mode in Maya (see Figure 4.3). In saturation sections, saturation values ranged from 24% to 86% with 270 in hue and 100% in value. A Hue of 270 is close to the colour purple that is considered a mild colour to avoid warm/cool colour bias for saturation test. In hue sections, hue varied from 0 to 331 with saturation and value in fixed values. In brightness sections, brightness ranged from 26% to 74% with saturation and hue in fixed values. See Table 4.2 for the settings of colour values in standard trials. To ensure the data is comparable, stimuli in each section were maintained equal contrast ratios to obtain psychophysical functions. Seven levels of colour contrast ratio between foreground and background in each section are fixed from 2.0:1, 2.1:1, 2.2:1, 2.3:1, 2.4:1, 2.5:1 and 2.6:1 by the algorism in the application Colour Contrast Analyzer developed by WAT-C. The standard stimulus is at level 2.3:1. (see Figure 4.4) Figure 4.3 HSV colour panel in Autodesk Maya 2011 Figure 4.4 The interface of WAT-C colour contrast analyser 91

106 4.3 Methods Table 4.2 Colour values in standard trials (F: Foreground object; B: Background object) Saturation standard Hue standard Brightness standard Ratio Object H S V Object H S V Object H S V 2.0:1 2.1:1 2.2:1 2.3:1 2.4:1 2.5:1 2.6:1 F % 100% F 0 100% 100% F % B % 100% B % 100% B % F % 100% F % 100% F % B % 100% B % 100% B % F % 100% F % 100% F % B % 100% B % 100% B % F % 100% F % 100% F % B % 100% B % 100% B % F % 100% F % 100% F % B % 100% B % 100% B % F % 100% F % 100% F % B % 100% B % 100% B % F % 100% F % 100% F % B % 100% B % 100% B % For a comfortable stereoscopic viewing, it is important to avoid divergence, where the left-eye image would appear more than 65mm (25inch) to the left of the right-eye image on the screen. The idea is that the maximum parallax must not exceed the human interpupillary distance 21 (IPD) when presented. To determine the maximum parallax, Dashwood (2011) suggested dividing the image horizontal resolution by screen width and then multiplying by IPD (65mm). In the experiment, the image horizontal resolution 1024 pixels divided by image projection width 1460mm, and multiply 65mm is 46 pixels. Looking at it another way, each pixel on the screen is 1.41mm wide by the calculation 65mm divided by 46 pixels. The parallax in the image stimulus is 18 pixels 21 Interpupillary distance (IPD) is the distance between the centre of the pupils of the two eyes. 92

107 4.3 Methods and the actual projected parallax on the screen is 25mm, which both are under the calculated maximum values. (see Figure 4.5 and 4.6) Figure 4.5 The width of actual projection size Figure 4.6 The parallax of actual projection Apparatus A polarised projection system was built for stimuli observations. Verrier (2009) pointed out current RealD 3D cinema utilises circularly polarised light to form stereoscopic image projection. The stimulus images were projected 93

108 4.3 Methods by two same model DLP projectors to a silver screen in a dim room, and observers were required to wear polarised glasses to fuse the left and right images at a fixed viewing distance (see Figure 4.7). Figure 4.7 Polarised projection utilised in this work The projection was run by two Optima EX531 DLP projectors with 1024*768 native resolution, 2600 ANSI Lumens and 2500:1 contrast ratio to project left and right images on the same screen synchronously. Polarised filters were set in front of projectors to polarise projection lights. A 100-inch silver screen was then used to display stimulus images. The actual projected image size on the screen was 1460mm x 1380mm. Viewers were asked to sit exactly 300 cm in front of the silver screen to warrant a comfort viewing experience (see Figure 4.8). 94

109 4.3 Methods Figure 4.8 Polarised projection and viewing distance. A Maxtron Dual Head 2 Go video splitter was connected between projectors and the computer to extend the output resolution to The computer employed for this experiment is a Macbook Pro notebook computer equipped with Intel Core 2 Duo 2.4 GHz CPU, 4GB DDR3 RAM and NVIDIA 9600GT graphic card. To superimpose left and right images onto a screen precisely, this study applied an coordinate image with the image size pixels to adjust the projection. Split by the video splitter, the coordinate image projected as two pixels images overlapped on the screen (see Figures 4.9, 4.10). Figure 4.9 Image used to align left and right projections 95

110 4.4 Projection Validation and Issues of Applying in Stereoscopic Projection Figure 4.10 Aligned projection 4.4 Projection Validation and Issues of Stereoscopic Projection Applying in The performance of laboratory instruments and equipment may change with time, either owing to fluctuations in the environment or owing to ageing of the optical or electronic components. Slow changes may not be obvious and can lead to errors in the results obtained. In addition, performance can be affected by repairs or replacement of modules or components. These potential instrumental errors can be controlled by carrying out regular preventative maintenance and calibration procedures. The performance validation of instruments and equipment should be based on traceable calibrators and validated standards, thus allowing equipment to be compared between different studies. The International Electrotechnical Commission (IEC) published an international standard, "IEC 61966", which defines methods and parameters for colour measurements and colour management for use in multimedia systems and equipment in The document "IEC " in the standard is the Part 6 that deals with front projection displays. The methods of measurement is designed to make possible the objective characterisation 96

111 4.4 Projection Validation and Issues of Applying in Stereoscopic Projection of colour reproduction of front projection displays corresponding to the intended colour. The measured results are intended to be used for the purpose of equipment-specific colour control in order to attain colour management in open multimedia systems. According to the standard, the arrangement of equipment for measurement should be shown as in Figure It incorporates a spectroradiometer or a non-contact colorimeter, depending on the characteristics to be measured. 1 Screen 2 Perfect reflecting diffuser 3 Front projection display 4 Spectroradiometer or colorimeter d Distance between screen and measuring instrument h Effective screen height H Height projected image L Distance from the screen Figure 4.11 Equipment arrangement for measurements. Figure is adapted from IEC (2005). The diagonal image size on the screen shall be set to the present size specified by the manufacturer. If no size is specified, it shall be set to 102cm. 97

112 4.4 Projection Validation and Issues of Applying in Stereoscopic Projection The height of front projection and the distance between the screen and the head of the front projection display shall be set to the present specified by the manufacturer. A perfect reflecting diffuser shall be set on the centre of effective screen area and the measured light shall be that reflected by a perfect reflecting diffuser. This study requires stereoscopic projection to present polarised stimuli to the observer. It requires two projectors both equipped with polarisers to project left and right stimuli simultaneously. IEC part 6 provides the standard to validate front projection display; however, several issues may raise if it applied to a stereoscopic projection system. The standard only supports CRT projectors that is colorimetrically controlled equipment using cathode ray tubes to present colour images with digital inputs for reference. The stereoscopic projection in this experiment is structured on two DLP projectors that apply micro mirror device to reflect single pixels in the resolution of the projected image. DLP technology has advantages of high light output from 5000 to ANSI lumens while CRT provides only overall 300 ANSI lumens. High refresh rate and contrast ratio in DLP projection also offer a more preferred stereoscopic viewing experience than CRT projection (McDowall and Bolas, 2002). Texas Instrument (2014) indicated there are 80 percent 3D movie theatres utilised DLP projection in the United States. To perform an industry-linked experiment, this study utilised DLP projection and therefore the IEC standard is not a fitting guideline to perform a referential validation. Secondly, the stereoscopic projection system requires polarisers equipped to externally filter the light and the viewer is required to wear a pair of polarising glasses to perceive the S3D images. Barco Projection System (2002) reported that the unpolarised light coming from each projector is diminished by the polarising filters and the efficiency is reduced to about 45%. Then the polarised image is viewed through polarising glasses that transmit polarised light at about 84% efficiency, resulting a final efficiency rate of approximately 38% (see Figure 4.12). IEC standard clearly 98

113 4.4 Projection Validation and Issues of Applying in Stereoscopic Projection indicates the equipment arrangement for measurement shall be shown in the Figure 4.11 and provides calculations of the normalised luminance level for different characteristics to be measure, but it doesn't include the condition of light transmission loss by equipping polarising filters. Moreover, characteristics measurement in the standard is based on single front projection display instead of dual projection display that required by a stereoscopic projection system and the introduction of dual simultaneous projection evaluation is not inclusive. It will not be possible to perform equipment validation to measure a dual simultaneous projection based on single projection guideline. Figure 4.12 The unpolarised light coming from each projector is externally polarised, which diminishes the brightness by more than half and reduces the efficiency to about 45%. Then the polarised image is viewed through stereo glasses that transmit the polarised light at about 84% efficiency, resulting in a final efficiency rate of approximately 38%. Finally, stereoscopic projection requires a silver high reflected screen to maintain the polarisation of the images to be presented to each eye. The metallic surface gives higher reflection than the common white ones. Additionally, the nature of polarised 3D projection requires the use of polarising filters, and the overall image is consequently less bright than a normal projection. A silver screen helps to compensate by reflecting more light back than a white one. IEC standard assumes the projection screen as the common white ones for evaluation and there is no supportive description to concern the screen light reflection. Problems may arise if the standard applied for a stereoscopic projection without verifying the intervention of screen light reflection. 99

114 4.5 Procedure IEC standard is introduced to validate projection devices basing on a projection technology that is currently and commonly in used today. The stereoscopic projection utilised in this study has different device conditions and it demands a more specific procedure to perform the device validation. Current information of stereoscopic projection evaluation is still fragmental and unaccountable. The validated outcome from the current standards and procedures would be invalid and meaningless. Consequently, in order to minimize the possible affection on the experiment outcomes caused from device moderation, this study applied brand new apparatus to avoid the issues of fluctuations and aging in the electronic components. Furthermore, the DLP projection display utilised in this study have been proved to have better chromaticity constancy than other projector technologies. Bakke, Thomas and Gerhardt (2009) indicated a comparison of chromaticity values of the ramps of red, green and blue between DLP, LCD and CRT projectors and DLP has higher constancy performance than other projectors. The result suggests that DLP projectors can provide a more stable colour performance and reduce the potential instrumental bias that is most concerned in device validation. 4.5 Procedure The method of constant stimuli was used to estimate thresholds for discriminating the stereoscopic depth perception by colour changes (see psychophysical procedure in Sections 3.11 and 3.12). A 2AFC procedure was applied, where observers saw standard and comparison stimulus images presented side by side on the screen and were required to choose the one they perceive more stereoscopic depth between foreground and background objects. The standard stimulus image was shown in each observation and randomly switched to the left and right. Observers were not told about standard and comparison stimulus, but simply instructed to choose the one with greatest stereoscopic depth. The instruction sheet can be found in Appendix B1. Observers made their decisions by pressing left or right arrow keys on the keyboard. The next trial came up automatically after 100

115 4.6 Results - Saturation a one second neutral gray colour screen interval throughout the experiment (see Figure 4.13). Trial A 1 Sec neutral gray colour interval Trial B Figure 4.13 Transition between trials 4.6 Results Saturation Saturation test comprised of standard and reversed trials. Seven observers participated this session, seven stimulus levels showed randomly, each level presented 50 times with a total 350 observations in each section. Figure 4.14 shows the psychometric functions of the standard trial; the saturation increasing in foreground. Figure 4.15 shows psychometric functions of the reversed trial; the saturation decreasing in the background. Seven levels of contrast ratio located on axis X while axis Y represents the proportion of correct responses. This experiment utilised SigmaPlot to find fitted sigmoid curves by running Sigmoid function analysis. The Sigmoid equation is as following. 101

116 4.6 Results - Saturation Figure 4.14 Psychometric function from 7 subjects in the standard trial. 102

117 4.6 Results - Saturation Figure 4.15 Psychometric function from 7 subjects in the reversed trial. 103

118 4.6 Results - Saturation In 2AFC procedure, the stimuli were presented side-by-side and the subject has to report which image stimulus contained more distance between foreground and background palettes. When the stimulus is below threshold, subjects are forced to guess and so the worst performance is 50% correct. The stimulus level where subjects were performing at 75% correct is taken to be the threshold. See Table 4.3 for the thresholds obtained in standard and reversed trials. Table 4.3 Psychophysical thresholds at 75% of psychometric function. Subject LD AM RE EN MJ GC JK Mean St. Dev. Standard Reversed In the standard trial, the range of the thresholds is between 2.36 to 2.45 with a mean In the reversed trial, the range of the thresholds is between 2.33 to 2.52 with a mean The results from Paired T-Test comparisons revealed that there is no statistically significant different between two trials. P>0.05 at 95% confidence level, fail to reject H0 (see Table 4.4). Table 4.4 T-test result concerning the standard and the reversed trials Test Hypothesis Two-tailed P value Conclusion H0: µstandard = µreversed vs. H1:µstandard >µreversed.81 Fail to reject H0 Table 4.5 show the means of subjects proportion of response in comparison stimulus in two sections. Figures 4.16 and 4.17 illustrate the averaged results from seven subjects and the results from linear regression show the slope (R²) in the standard trial is 0.96 and the slope in the reversed trial is

119 4.6 Results - Saturation Table 4.5 Proportion of response in comparison stimulus Level 2.0:1 2.1:1 2.2:1 2.3:1 2.4:1 2.5:1 2.6:1 Standard Reversed Standard Reversed Figure 4.16 Linear regression from the standard trial. Figure 4.17 Linear regression from the reversed trial. 105

120 4.6 Results - Hue The coefficients in linear regressions indicate the slopes in two trials are significant different from 0, which means there is a significant relationship between colour contrast of saturation and S3D perception as the test hypothesis β 0 and p<0.05 at 95% confidence level. (see Table 4.6). Table 4.6 T-test results of slopes Trials Hypotheses P value Conclusion Standard H0: β = 0 vs. H1: β Reject H0 Reversed H0: β = 0 vs. H1: β Reject H0 An univariate analysis of variance was then performed on the comparison concerning the slopes. The result showed there is no significant difference between the variances in standard and reversed trials. 1.2 as p>0.05 at 95% confidence level. (see Table 4.7). Table 4.7 Univariate analysis of variances Source Sum of Squares df F Sig. Standard * Reversed Hue Hue test comprised of the standard and the reversed trials. Colours were arranged warm hue in foreground with cool hue in background in the standard trial; cool hue in foreground with warm hue in background in the reversed trail. Figures 4.18 and 4.19 show the psychophysical function of Hue changes. Thresholds are expressed as the amount of contrast ratio that was applied to the ratio of Hue change. Data for the 75% threshold for the seven observers are shown in Table

121 4.6 Results - Hue Figure 4.18 Psychometric functions in the standard trial. 107

122 4.6 Results - Hue Figure 4.19 Psychometric functions in the reversed trial. 108

123 4.6 Results - Hue Table 4.8 Psychophysical thresholds at 75% of psychometric function. Subject LD AM RE EN MJ GC JK Mean St. Dev. Standard Reversed In the standard trial, the range of the thresholds is between 2.37 to 2.46 with a mean In the reversed trial, the range of the thresholds is between 2 to 2.48 with a mean The P-value from Paired T-Test comparisons is greater than 0.05, which indicated there is no statistically significant different between two trials. P>0.05 at 95% confidence level, fail to reject H0. (see Table 4.9). Table 4.9 T-test result concerning Standard and Reversed trials. Test Hypothesis Two-tailed P value Conclusion H0: µstandard = µreversed vs. H1:µstandard > µreversed.20 Fail to reject H0 Table 4.10 shows the means of subjects proportion of response in comparison stimulus in two sections. Figures 4.20 and 4.21 are the averaged results from seven subjects and the results from linear regression show the slope (R²) in the standard trial is and the slope in the reversed trial is Table 4.10 Proportion of response in comparison stimulus. Level 2.0:1 2.1:1 2.2:1 2.3:1 2.4:1 2.5:1 2.6:1 Standard Reversed

124 4.6 Results - Hue Standard Reversed Figure 4.20 Linear regression of the standardd trial. Figure 4.21 Linear regression of the reversed trial The coefficients in linear regressions indicate slopes in two trials are both significant different from 0, which means there is a significant relationship between contrastt of hue and S3D perception as β 0 at p<0.05 at 95% confidence level. (see Table 4.11). 110

125 4.6 Results - Brightness Table 4.11 T-test results of slopes Sections Test Hypotheses P value Conclusion Standard H0: β = 0 vs. H1: β Reject H0 Reversed H0: β = 0 vs. H1: β Reject H0 An univariate analysis of variance was then performed on the comparison concerning the slopes. The result showed there is no significant difference between the variances in standard and reversed trials as p>0.05 at 95% confidence level. (see Table 4.12). Table 4.12 Univariate analysis of variances Source Sum of Squares df F Sig. Standard * Reversed Brightness Brightness test comprised of the standard and reversed trials. Higher brightness in foreground with lower brightness in background was arranged in the standard trial. Lower brightness in foreground with higher brightness in foreground was arranged in the reversed trial. Figures 4.22 and 4.23 show the psychophysical function of Brightness tests. Psychometric function which shows the relationship between the percentage that a stimulus is perceived and the corresponding stimulus intensity. Threshold is then obtained at 75% of psychometric function. The mean score and standard deviation for two sections are shown in Table

126 4.6 Results - Brightness Figure 4.22 Psychometric functions in the standard trial. 112

127 4.6 Results - Brightness Figure 4.23 Psychometric functions in the reversed trial. 113

128 4.6 Results - Brightness Table Psychophysical thresholds at 75% of psychometric function. St. Subject LD AM RE EN MJ GC JK Mean Dev. Standard 2.50 n/a n/a 2.59 n/a n/a n/a n/a n/a Reversed In the standard trial, only two out of seven subjects have reached 75% threshold. In the reversed trial, the range of the thresholds is between 2.37 to 2.56 with a mean Due to five subjects couldn t reach the 75% threshold in the standard trial, it is therefore not able to perform a T-test comparison between the standard and the reversed trials. However, it is very interested to look at the slopes comparison to see if there was a difference in between. Table 4.14 shows the means of subjects proportion of response in comparison stimulus in two sections. Figures 4.24 and 4.25 show the averaged results from seven subjects and the results from linear regression. Table 4.14 Proportion of response in comparison stimulus Level 2.0:1 2.1:1 2.2:1 2.3:1 2.4:1 2.5:1 2.6:1 Standard Reversed

129 4.6 Results - Brightness Standard Reversed Figure 4.24 Linear regression in the standardd trial. Figure 4.25 Linear regression in the reversed trial. The coefficients in linear regressions indicate slopes in two trials are significant different from 0, which means there is a significant relationship 115

130 4.7 Summary of Results between contrast of brightness and S3D perception as β 0 at p<0.05. (see Table 4.15) Table 4.15 T-test results of slopes Sections Hypotheses P value Conclusion Standard H0: β = 0 vs. H1: β Reject H0 Reversed H0: β = 0 vs. H1: β Reject H0 An univariate analysis of variance was then performed on the comparison concerning the slopes. The result showed there is a significant difference between the variances in the standard and reversed trials as p<0.05. (see Table 4.16). Table 4.16 Univariate analysis of variances Source Sum of Squares df F Sig. Standard * Reversed Summary of Results Discrimination of stereoscopic depth changes in different colour arrangements The data summarised in Table 4.6, 4.11 and 4.15 shows that the observers are able to discriminate depth changes with different levels of colour contrast as the coefficient results are significant different from 0. The results indicate that colour contrast in hue, saturation and brightness does influence stereoscopic depth perception. These findings are cooperative with Dresp and Guibal s (2004) conclusion that the colour of objects has a considerable effect on depth perception in the visual field. The findings in this work further 116

131 4.7 Summary of Results indicate that the influence of colour exists even under polarised projection viewing condition Significance and Comparison of Thresholds Obtained In the saturation section, both standard (SS) and reversed (SR) trials achieved thresholds at 75% and the limen of contrast ratio is between 2.00 to The thresholds from two trials show a similarity which observers have the averaged threshold in saturation section at 2.40 contrast ratio. The psychometric functions show stable sigmoid curves for each viewer (see SS curve in Figure 4.24). Only certain viewers (EN and RE) showed larger thresholds for the standard trial than the reversed trial. The ANCOVA result in Table 4.7 shows there is no significant variation between observers. In the hue section, the ANCOVA test shows no significant difference between standard (HS) and reversed (HR) trials; however, the averaged thresholds have larger difference than the saturation section. Viewer AM shows a total reversed sigmoid curve in the trial HR which still achieved 75% threshold at 2.0 contrast ratio while other viewers still remained stable curve shapes (see Figure 4.24). This resulted in a lower mean in HR but didn't cause a significant deviation in the psychometric functions. If we remove AM's data, the averaged threshold from the six viewers would be 2.38 and this led to a more homogeneous result with HS trial that has the averaged threshold at It is speculated that a reversed thresholds obtained by AM might reflect that warm colour arranged in the background can still be perceived nearer to the observer than the cool colour in the foreground. Hue as a psychological depth cue can be more dominant than binocular disparity as a physiological depth cue in some cases. Bülthoff and Mallott (1988) indicated that depth cues veto one another when there is conflicting information, there is weighting among depth cues and a prioritisation of which cures are ignored when in conflict. The topic of depth cues conflict is not the major focus in this study however the result from AM can be described as a deviation from depth cues dominance. In the brightness section, the standard trial (BS) shows more flat psychometric functions curves in all viewers except viewer LD (see Figure 2.24). Only LD and En achieved 75% threshold at 2.50 and 117

132 4.7 Summary of Results 2.59 contrast ratio, however their curves also show deviated patterns compared with other five trials that stay in a sigmoid shape. Apart from viewer MJ's nearly horizontal curve, all viewers curves in BS trial still show a small ascendant drift from low to high contrast ratios. On the other hand, the reversed (BR) trial showed consistent sigmoid shapes in each viewer and achieved an averaged threshold at The ANCOVA test then consequently indicates a significant difference between BS and BR trials. See Figure 4.26 for the psychometric function comparisons. 118

133 4.7 Summary of Results Figure 4.26 Psychometric function comparisons 119

134 4.7 Summary of Results Comparison of Standard and Reversed Colour Arrangements From the thresholds obtained in standard and reversed experiments in Tables 4.3 and 4.8, it can be seen that observers show similar discrimination of colour contrast changes in saturation and hue applied to stereoscopic depth perception. There is no significant difference in the results of T-test in Tables 4.4 and 4.9 between the thresholds obtained in standard and reversed trials. The results of slope comparison shown in Tables 4.7, 4.12 and 4.16 indicate there is no significant difference between the standard and reversed trials in saturation and hue sections but there is a significant difference between standard and reversed brightness trials. Egusa (1983) and Bailey et al. (2006) concluded that high saturated objects appear more advance than low saturated objects. The results from this experiment responds a coherent conclusion under a polarised projection viewing condition. This experiment additionally ran a background decrease trial and found it is equally efficient to increase saturation in foreground and decrease saturation in background. The results from hue correspond with Dengler and Nitschke s (1993) findings that long-wavelength colours appear in the front. However, the results from reversed trail show no significant difference with the standard trial. The similarity between standard and reversed trials in saturation and hue sections suggests an inference that colour contrast is the dominant depth cue over colour attribute itself in stereoscopic depth perception. This finding replies Farne s (1977) conclusion that stimulus have lower contrast with the background appeared farther. It is therefore this finding can suggest that colour contrast is a more influential consideration than colour arrangements of hue and saturation for foreground and background in stereoscopic depth perception. 120

135 4.7 Summary of Results The results from psychophysical function in brightness section show that subjects have poor ability to discriminate the depth changes in the standard trial. Only two out of seven subjects reached 75% threshold. The slope difference between standard and reversed trials in brightness section indicates that higher brightness in foreground resulted less depth discrimination than higher brightness in background. The finding can respond to the conclusion from Camgöz, Yener and Güvenç (2004), which mentioned brightness is the most dominant attribute in depth perception. In the case of this work, the brightness is more dominant than colour contrast. It is therefore this finding can suggest that brightness attribute is more dominant then colour contrast in stereoscopic depth perception. The influence can be positive (reversed trial) and negative (standard trial) for stereoscopic depth perception and the arrangement of lower brightness in foreground is more effective for preferred stereoscopic depth perception. More discussion about the results of psychophysical thresholds can be found in Chapter

136 5.1 Introduction CHAPTER 5 Evaluating Saturation, Hue and Brightness in Stereoscopic 3D Perception by Image Quality Assessment in CGI 3D Scenes 5.1 Introduction Chapter 4 demonstrated that observers are able to discriminate small changes in stereoscopic depth when altering colour values in saturation, hue and brightness. To avoid any possible distraction, the psychophysical experiments in Chapter 4 applied colour palettes to present the stimulus images. The discrimination thresholds showed that based on fixed contrast ratios there are no significant difference to change colours in foreground and background palettes except brightness. The results in Chapter 4 conclude that colour changes do influence stereoscopic depth perception from a human factor aspect. In order to strengthen the validity of the result, this chapter performs stereoscopic depth quality assessments based on ITU-R Recommendations BT , which constitute a set of international technical standards developed by the Radiocommunication Sector (formerly CCIR) of the ITU (International Telecommunication Union). This chapter discusses the colour changes from an image quality viewpoint by performing subjective stereoscopic depth quality assessments, and applies practical computer-generated 3D scenes as stimulus images. The scenes are designed associated to filmmaking, gaming and commercial images, which are the fields that S3D content mostly applied. The colour arrangements in scenes are same with Chapter 4 that described in Table 4.1. The depth quality evaluation in this chapter was expected to provide the verification to the results in Chapter 4 coherently, therefore, the results from 122

137 5.2 Aims and Predictions psychophysical test of colour palettes are referential to compare with the results from this chapter. 5.2 Aims and Predictions The aim of this experiment is to evaluate the depth quality by changing colour values in CGI 3D scenes. A subjective human-based trial is performed based on the ITU-R BT Recommendation to assess the depth quality by six different colour settings. Based on the results from the psychophysical experiment in Chapter 4. The following predictions are tested: 1. Higher saturation in foreground scene has no significant difference with lower saturation in background scene on stereoscopic depth quality. 2. Warm hue in foreground with cool hue in background scene has no significant difference with cool hue in foreground with warm hue in background scene on stereoscopic depth quality. 3. High brightness in foreground scene has weaker stereoscopic depth quality than low brightness in foreground scene. 4. Contrast is more dominant in stereoscopic depth perception than saturation and hue, but weaker than brightness. 5.3 Methods The procedure applied in depth quality assessment follows the Stimuluscomparison Method from ITU-Recommendation described in Chapter 3, Section Test images generated from six colour designs are presented three times randomly side by side with a standard image to the observer. Subjects are required to give a score of stereoscopic depth quality on the 123

138 5.3 Methods test image based on the comparison with the quality of the standard image. Six different colour arrangements are applied in three different CGI 3D scenes Observers Nineteen subjects, nine male and ten female, took part in the experiment, ages ranged from 23 to 34 with a mean of 28 years. Subjects were not aware of the purpose of the experiment and they were all non-expert in stereoscopic imaging. All the participants met the minimum criteria of stereoacuity test at 40 sec-arc and passed the Ishihara Colour Blindness Test. All participants were postgraduates recruited from within the university and varied in nationalities and culture backgrounds. Five British, four Chinese, four Taiwanese, three Korean, one Indian, one Greek and one Nigerian had participated the experiment. Observers culture background was not considered as an intervening factor in this experiment as described in Section Procedure The experiment divided into three sections, saturation, hue and brightness sections. Participants were allowed to have 5 minutes break between sections. All of them were required to read an instruction before the trials. The instruction sheet can be found in the Appendix B2. The judgement request for each observation stated in the instruction Select the image you perceive higher stereoscopic depth quality. The instructions contained example images to illustrate the foreground and background objects in the scene and explained the definition of stereoscopic depth quality. The stereoscopic 3D images were projected on a screen and each observer was required to sit at a fixed viewing distance to the screen. The viewers were 124

139 5.3 Methods asked to record his/her results by verbally report the score to the experimenter. Each test image lasts up to 30 seconds, the experimenter operated to next trial when the viewer reported the score, followed by a onesecond blank interval of gray between trials. All participants were given the chance to ask question before, during and after the trial and understand they were free to withdraw from the experiment at any time. The two vision tests took about 5 minutes and the test session last about 20 minutes. Each participant finished a total 54 observations and received a gift as reward. The reference stimulus image was showed in each trial. Observers were told about reference and comparison stimulus and asked to rank the stereoscopic depth quality. Observers made their decisions by pressing 1 to 9 number keys on the keyboard. The trial switches to next one after a one second neutral gray colour screen interval throughout the experiment. (see Figure 5.1). Figure 5.1 Illustration of gray interval between trials Scoring Scale In each trial the test image is rated on a scale of Excellent, Good, Fair, Poor and Bad. These terms categorise the five different levels and they are the same as those normally used in the ITU-R recommendation. The terms are associated with the value intervals of 9 to 7, 7 to 5, 5 to 3, and 3 to 1, respectively. (see Figure 5.2) The viewers are asked to score the depth quality of each test image based on the comparison with the reference 125

140 5.3 Methods image presented side by side which is defined at score 5. The vertical scale was displayed on the instruction paper and divided into four equal lengths. 9 Excellent 8 7 Good 6 5 Fair 4 3 Poor 2 1 Bad Figure 5.2 Nine points scoring scale based on ITU-R BT Apparatus The apparatus were arranged as described in Chapter 4, Section Stimuli All stimuli designed and rendered in Autodesk Maya See general stimulus images design method in Chapter 3, Section 3.5. Three different scenes are designed and associated to saturation, hue and brightness. Each scene contains a main foreground object with the colour value set differently from the background colour. Observers judged the stereoscopic depth quality harmonised from foreground main object and background objects. 126

141 5.3 Methods Scene Design Three different composition scenes were associated to animation filmmaking, commercial image and gaming scenes, respectively. The settings of the virtual stereo camera in each scene are consistent; same values of parallax, interaxial separation and focal length, which associated to stereopsis that are set in each scene to avoid different stereo convergences. Images were captured by the stereo camera rig into left and right images for stereoscopic presentation. Scene 1 is a forest scene, composed of colourful mushrooms, grasses, a tree trunk, a dark sky and light particles on the ground. The main foreground object in is a mushroom. See Figure 5.3 for the scene in Maya and the final rendered scene. Scene 2 is a commercial product scene, composed of a perfume bottle in the foreground and cosmetic objects around the background. A dramatic light is set. A spotlight is set in the right side of the scene to emphasis the outline and shadow of the object (see Figure 5.4). Scene 3 is a fountain set in the foreground of an old time street, a brick building with doors and windows set in the background under sunny daylight. (see Figure 5.5) Figure 5.3 Scene 1- The left-hand image is the scene in Maya from a perspective viewport. The right-hand image is the final rendered image. 127

142 5.3 Methods Figure 5.4 Scene 2 - The left-hand image is the scene in Maya from a perspective viewport. The right-hand image is the final rendered image. Figure 5.5 Scene 3 - The left-hand image is the scene in Maya from a perspective viewport. The right-hand image is the final rendered image Colour Values References images in the assessments are adjusted only in hue to a colour value at 270 (see Figure 5.7). The purpose is that the purple colour is considered a mild colour to avoid warm/cool colour bias for all trials (Caponigro, 2011). Figure 5.6 illustrates the colour wheels with hue values of warm and cool sides. 128

143 5.3 Methods Figure 5.6 Colour wheels with hue values and cool-warm sites. Figures adapted from Caponigro (2011). Figure 5.7 Reference images for scene 1 (top left), scene 2 (top right) and scene 3 (bottom right). Based on the colour values of reference images, the colour values of test images were adjusted in foreground and background areas by HSV colour panel in Adobe Photoshop. The main foreground object is lassoed and colour changed separately with rest area of the image. For higher saturated foreground test images, saturation value controlled at a range from 60% to 90% in foreground object and 40% to 60% in background area. For lower 129

144 5.3 Methods saturated background test images, saturation value controlled at a range from 40% to 60% in foreground object and 10% to 40% in background area. See Table 5.1 for the colour value ranges and stimulus images. 130

145 5.3 Methods Table 5.1 Colour values in test images (F: Foreground, B: Background) Higher Saturated Foreground F: 60% to 90% B: 40% to 60% Lower Saturated Background F: 40% to 60% B: 10% to 40% Warm hue in foreground with cool hue in background F: 0 to 30 B: 180 to 210 Cool hue in foreground with warm hue in background F: 180 to 210 B: 0 to 30 Brightness test images F: 60% to 90% B: 10% to 40% Brightness test images F: 10% to 40% B: 60% to 90% Scene 1 Scene 2 Scene 3 131

146 5.4 Results - Saturation 5.4 Results Saturation Figure 5.8 shows the box plot from all nineteen subjects. The score range in the standard trial is from 4.56 to 6.67, and 3.44 to 6.44 in the reversed trial. The 25th percentile and the 75th percentile in the standard trial is at 5 and while the reversed trial is at and The medians are the same in both trials at Each observer s scores are averaged and shown in Table 5.2. Figure 5.8 Box plot in the saturation section. As shown in Table 5.2, only four subjects out of nineteen in the standard trial have lower scores than standard score 5 which correspond with the term Fair in ITU s grading score. In the reversed trial, the mean is a little bit higher than Fair ; however, there are nine subjects out of nineteen have lower score than the standard score 5. The means in two sections are both above the standard score. 132

147 5.4 Results - Saturation Table 5.2 Averaged scores from scene 1, scene 2 and scene 3 in the saturation section. Subject Standard Reversed Subject A Subject B Subject C Subject D Subject E Subject F Subject G Subject H Subject I Subject J Subject K Subject L Subject M Subject N Subject O Subject P Subject Q Subject R Subject S Mean Std Deviation T-Test was performed to examine the difference between the means and standard score. The mean in the standard trial is significantly different from 5, which means there is a positive significant relationship between higher saturated foreground arrangement and stereoscopic depth perception as the 133

148 5.4 Results - Hue test hypothesis µstandard 5 and p<0.05 at 95% confidence level. (see Table 5.3). On the other hand, the result from the reversed trial indicates that there is no significant difference between the mean and standard score. Table 5.3 T-test concerning means and the standard score 5. Sections Hypotheses P value Conclusion Standard H0: µstandard = 5 vs. H1: µstandard 5.01 Reject H0 Reversed H0: µreversed = 5 vs. H1: µreversed 5.48 Fail to reject H0 A two-tailed T-Test was then performed to examine the significance between the standard and the reversed trials. The result indicates that there is no statistically significant between two arrangements with p>0.05 at 95% confidence level, fail to reject H0. (see Table 5.4). Table 5.4 T-test concerning the standard and the reversed trials. Test Hypothesis Two-tailed P value Conclusion H0: µstandard = µreversed vs. H1: µstandard > µreversed.27 Fail to reject H Hue Figure 5.9 illustrates the box plot of the standard trial that warm hue was arranged in foreground with cool hue in background and the reversed trial that cool hue was arranged in foreground with warm hue in background. Means and standard deviation are shown in Table

149 5.4 Results - Hue Figure 5.9 Box plot in hue section. There are eight subjects out of nineteen in the standard trial have lower scores than the standard score 5. The lowest averaged score is 3.44, which correspond between the terms Fair and Poor in ITU s grading score. The mean 5.08 is very close to Fair. In the reversed trial, there are only three subjects out of nineteen have lower score than standard score

150 5.4 Results - Hue Table 5.5 Averaged scores from scene 1, scene 2 and scene 3 in the hue section. Subject Standard Reversed Subject A Subject B Subject C Subject D Subject E Subject F Subject G Subject H Subject I Subject J Subject K Subject L Subject M Subject N Subject O Subject P Subject Q Subject R Subject S Mean Std Deviation The T-Test result showed the mean in the standard trial has no significant different with the standard score, µstandard = 5 and p>0.05, which mean the stereoscopic depth quality stay similar when arranging warmer colour in the foreground. However, the result from reversed trial indicates that there is a significant difference between the mean and standard score. µreversed 5 and p<0.05. (see Table 5.6) 136

151 5.4 Results - Brightness Table 5.6 T-test concerning means and the standard score 5. Sections Hypotheses P value Conclusion Standard H0: µstandard = 5 vs. H1: µstandard 5.68 Fail to reject H0 Reversed H0: µreversed = 5 vs. H1: µreversed 5.00 Reject H0 A two-tailed T-Test was then performed to examine the significance between the standard and the reversed sections. The result indicates that there is no statistically significant between two arrangements as p>0.05 at 95% confidence level, fail to reject H0. (see Table 5.7) Table 5.7 T-test concerning the standard and the reversed trials. Test Hypothesis Two-tailed P value Conclusion H0: µstandard = µreversed vs. H1: µstandard >µreversed.27 Fail to reject H Brightness Figure 5.10 illustrates the box plot of brightness section which comprising the standard and the reversed trials. High brightness was arranged in foreground in the standard trail and low brightness was arranged in foreground in the reversed trial. The scores from nineteen subjects are averaged and the means and standard deviation are shown in Table

152 5.4 Results - Brightness Figure 5.10 Box plot in the brightness section. There are thirteen subjects out of nineteen in the standard trial have lower scores than score 5. The range of averaged scores is from 3.78 to 5.56, which correspond with the terms Poor in ITU s grading score. The mean µstandard, 4.55, is the lowest in six trials. In the reversed trial, there are twelve subjects out of nineteen have lower score than standard score 5. The range of averaged score is from 3.78 to 6 with the mean

153 5.4 Results - Brightness Table 5.8 Averaged scores from nineteen subjects in the brightness section. Subject Standard Reversed Subject A Subject B Subject C Subject D Subject E Subject F Subject G Subject H Subject I Subject J Subject K Subject L Subject M Subject N Subject O Subject P Subject Q Subject R Subject S Mean Std Deviation A T-test concerning the means and standard score in the standard trial showed a significant difference with standard score, µstandard 5 and p<0.05, which means there is a significant relationship between high brightness foreground arrangement and stereoscopic depth perception. However, the result from the reversed trial indicates that there is no significant difference between the mean and standard score as µreversed = 5 and p>0.05, which mean 139

154 5.5 Summary of Results the stereoscopic depth quality stay similar when arranging brighter colour in the background. (see Table 5.9) Table 5.9 T-test concerning means and the standard score 5. Sections Hypotheses P value Conclusion Standard H0: µstandard = 5 vs. H1: µstandard 5.00 Reject H0 Reversed H0: µreversed = 5 vs. H1: µreversed 5.43 Fail to reject H0 A two-tailed T-Test was then performed to examine the significance between the standard and the reversed trials. The result indicates that there is a statistically significant between two brightness arrangements as p<0.05 at 95% confidence level, reject H0. (see Table 5.10) Table 5.10 T-test concerning the standard and the reversed trials. Test Hypothesis Two-tailed P value Conclusion H0: µstandard = µreversed vs. H1: µstandard > µreversed.05 Reject H0 5.5 Summary of Results This chapter performed a subjective human-based assessment to evaluate depth quality in different coloured CGI stereoscopic 3D scenes which aim to verify the conclusion in Chapter 4 coherently. As showed in Table 5.11, only the means of standard and reversed trials in brightness section fell below 5 that corresponded with the term Fair in ITU s grading scale. Colour arrangements with saturation and hue can provide positive effect on stereoscopic depth perception. The finding corresponds to the conclusion from Camgöz, Yener and Güvenç (2004), which indicated brightness is the most dominant attribute in depth perception. In the case of this assessment, brightness dominates the depth perception in a negative way concerning 140

155 5.5 Summary of Results stereoscopic depth enhancement; however, the reversed trail performed higher depth quality than the standard trial. This result replies to the conclusion in Chapter 4 that the arrangement of low brightness in foreground is more effective than high brightness in foreground. Table 5.11 Means of score in 6 trials Sections Saturation Hue Brightness Trials Standard Reversed Standard Reversed Standard Reversed Means The T-test results in comparison between standard and reversed trails showed in Table 5.4, 5.7 and 5.10 indicate that increasing saturation in foreground has equally depth quality with decreasing saturation in background in stereoscopic depth perception. Warm hue in foreground with cool hue in background has similar quality to cool hue in foreground with warm hue in background. Low brightness in foreground has higher depth quality than high brightness in foreground. These findings all correspond to the results in Chapter 4 described in Section 4.6. Predictions described in Section 5.2 are confirmed in Table More discussion about the results can be found in Chapter

156 5.5 Summary of Results Table 5.12 Confirmation of the predictions Hypothesis High saturation in foreground scene has no significant difference with low saturation in background scene on stereoscopic depth quality. Warm hue in foreground with cool hue in background scene has no significant difference with cool hue in foreground with warm hue in background scene on stereoscopic depth quality. High brightness in foreground scene has weaker stereoscopic depth quality than low brightness in foreground scene. Contrast is more dominant in stereoscopic depth perception than saturation and hue, but weaker than brightness. Confirmation True True True True 142

157 6.1 Saturation and Hue CHAPTER 6 General Discussion The experiments described in this thesis investigate colour arrangements in stereoscopic depth perception using polarised 3D projection that is intended to be more correspond to the applications at current 3D cinema than simple stereoscopes used in previous related experiments. This has been undertaken using stimuli composed of a foreground colour with a background colour series, which generate the parallax from left and right images rendered from computer virtual scenes. The colour changes were applied to the stimuli incorporated changes to colour contrast in saturation, hue and brightness that are the major colour adjustments in cinematic imaging design. In addition, the S3D depth quality evaluation in Chapter 5 incorporated practical scene designs was measured under ITU-R Recommendations; this represents a closer approximation to conditions that are experienced in colour adjustments practice for computer-generated imagery, which tend to require the analysis of S3D depth quality under the procedure of international standard. 6.1 Saturation and Hue The body of data obtained from the studies described in this thesis provides the evidence that colour is an influential depth cue in depth perception. These results are consistent with works on depth perception analysis that are based on monocular depth judgements (e.g. Dresp and Guibal, 2004). The results also respond to the works that claimed contrast is an important cue for perceiving the depth of an object (e.g. Payne 1964; Ichihara et al. 2007). This proposal extends the work advanced by O'shea et al. (1993) which suggested that contrast adjusted by varying luminance is an effective depth cue, to a situation where this process can be applied to contrast adjusted by saturation, hue and brightness, and to a more complex polarised 143

158 6.1 Saturation and Hue 3D viewing condition other than the monocular viewing on a computer screen. In addition, the data suggests that saturation is influential in S3D depth perception, but the situation of the high saturated object appears nearer than the low saturated one (e.g. Luckiesh 1918; Egusa 1983; Bailey et al. 2006) is not determined in S3D viewing. The standard and reversed trials in the psychophysical experiments provide similar thresholds with no significant difference. Evidence from the depth quality evaluation in Chapter 5 corresponds a coherent conclusion that there is no considerable difference to arrange saturation between foreground and background objects to determine S3D depth perception. The same result was evident in the hue section, the phenomenon of chromatic aberration (Sundet 1978) and the claim that long-wavelength colours appear nearer than short-wavelength colours (Dengler and Nitschke 1993) are not applied in S3D depth perception. Different hues do affect viewers depth judgement, but the psychophysical thresholds have no significant difference between the standard and reversed trials and the results from the depth quality evaluation supports the same conclusion. Conclusively, saturation and hue are effective moderators when considering colour design for S3D perception, but this postulate is conditional upon the manner of the background colour. In this study, the foreground and background colours in the experiments are controlled by fixed colour contrasts in order to acquire comparable data to analyse between different sections. Therefore, the insignificant differences between the standard and reversed trials in saturation and hue can be attributed to the dominance of colour contrast. The inference responds to the study indicated that the saturation difference between the stimulus and the surrounding bears importance for depth perception (Mount et al. 1956), and the claim mentioned that contrast is an important cue for perceiving the depth of an object (e.g. Payne 1964; Ichihara et al. 2007). Adjacent colours referring to stimuli-surrounding relation appear to be more crucial when considering depth perception other than merely concern of colour saturation and hue. 144

159 6.2 The Controversy of Brightness 6.2 The Controversy of Brightness Saturation and hue are named as the secondary attributes of colour in judgment of nearness, whereas the brightness is the most dominant attribute (Camgöz, Yener and Güvenç 2004). The results from this study suggest that brightness is certainly more dominant than saturation and hue but not in a desired manner concerning S3D colour design. In the brightness section, the viewers showed poorer ability to discriminate the changes in the standard and reversed trials than saturation and hue sections. The dominance of brightness overpowers the depth cue of colour contrast and interferes viewers judgement of depth perception. Studies investigated brightness and contrast as indicators to distance indicated that a bright area next to a dark area appears brighter than it really is. Brightness may cause the simultaneous contrast which has the effect of darkening the dark colour and lightening the light colour more (Grandis 1986). This effect happened in this study when adjusting contrast ratios for brightness stimuli. To remain the fixed contrast ratios, brightness value can only be increased or decrease by a relatively smaller volume than saturation and hue due to it has high intensity to influence contrast. Therefore, the overall lower discrimination performance in the brightness section can be understood by the small changes presented. It will be interesting to see if we apply intensity volume as the unit for the experiments, such as the volume percentage in Adobe Photoshop, instead of contrast ratios, brightness may perform a higher discrimination than saturation and hue. The results in brightness also reveal a difference between the standard and reversed trials. Bright foreground with dark background showed a low discrimination while dark foreground with bright background still remained a stable perceiving in changes. The depth quality evaluation suggested a similar result to confirm the outcome. The result contradicts the well-known concept according to which the brighter object should in any case be perceived as the nearer (Payne 1964; Michel 1996). The works of stimulisurrounding relation indicated that in different levels of brightness, the one having higher contrast with its background is perceived as the nearer, and 145

160 6.2 The Controversy of Brightness the darker ones appear nearer to the viewer when viewing with the off-white background (Farnè 1977). It can be understood that contrast still play an important factor in brightness and the possible equivalent between the standard and reversed trials in this study, but what effect can explain the dark foreground stimulus may have higher effectiveness in depth perception than bright foreground stimulus under a equal contrast setting? Works discussed background brightness pointed that when colours are seen in front of a black background, the long-wavelength colours appear nearer; when seen in front of a white background, the depth perception of colours are reversed (Dengler and Nitschke 1993). In that sense, the brightness of a stimulus affects depth perception of colour as well. This point of view may explain that low brightness background enhances the nearness judgment for long-wavelength colours while high brightness background helps shortwavelength colours to stand out. The stimuli in the brightness section in this study were designed purely considering brightness value without involving saturation and hue in order to eliminate possible bias in the data. Therefore, it is interesting to see if brightness can be measured in the wavelength spectrum. The Helmholtz-Kohlrausch (HK) effect describes that the brightness of a stimulus is not a simple representation of luminance since the brightness of equally luminant stimuli changes with their relative saturation, and with shifts in the spectral distribution of the stimulus (Helmholtz 1924; Pridmore 2007). An important aspect of the HK effect is the asymmetry in the wavelength-dependency of the effect. Specifically, short-wavelength light, such as blue, appears brighter than long-wavelength light, such as red (Ayama and Ikeda 1998). Therefore, it can be explained that in the standard trial, a bright foreground was combined with a dark background, which resulted a short-wavelength light object was perceived receded combining with a low brightness background, and this effect confused the viewer when judging depth perception. However, there is still a gap on depth perception of brightness in terms of the stimuli and its background. Some works claimed that the response to colour cannot be separated from the reaction on brightness due to these components are closely connected with each other (e.g. Weiss et al. 1943). 146

161 6.3 The Dominance of Contrast 6.3 The Dominance of Contrast Of the various colour attributes proposed to account for the investigation of S3D depth perception, five trials out of six in the psychophysical experiments showed that contrast has successfully dominant in nearness judgment. The results correspond to the studies indicated that low contrast stimuli appeared farther than high contrast stimuli (Schor and Howarth 1986). This study further provides the evidence that contrast has its effect on stereoscopic viewing condition (Rohaly and Wilson 1993). The results also respond to the stimuli-surrounding relation that when a colour differs from the background more, and it stands farther away from the background; one of the two colours has more similarity to the background will appear more distant than the other (Payne 1964). The results from depth quality evaluation in this study again correspond the claim that contrast creates some illusions on the appearance of coloured surfaces in the manner of depth perception (Ichihara et al. 2007). Most studies in contrast as a depth cue demonstrated the nearness judgment in a monocular viewing condition and discussed the depth effects of contrast in the aspect of a simulation of aerial perspective (Egusa 1983; (Farnè 1977; Mount et al. 1956; Ross 1968; Troscianko et al. 1991). These investigations support that contrast can act as a pictorial cue to depth. However, few investigations have proven that contrast has its influence on the mechanism of stereopsis like this study did. Fry and Bridgman (1942), Schor and Howarth (1986) have studied this effect under stereoscopic viewing and indicated that contrast is dominant in depth perception. Rohaly and Wilson (1991) concluded that low-contrast object will produce a weaker responses to the disparity-tuning curves of the "near" and "far" that they developed than high-contrast object at the same distance. This may lead to the claim by Edwards (1955) that colour has no effect on quality of depth. The results from this study propose that colour will only be considered in stereoscopic depth quality when there is a contrast formed from its background colour. Specifically, contrast is dominant in depth perception, 147

162 6.4 Limitations but colours combinations can be considered to obtain the desired contrast for a preferred stereoscopic depth perception. 6.4 Limitations Several of the potentially limiting factors associated with experiments performed in this thesis were addressed during the course of the studies that have been undertaken. One of the major early issues was the strategies used for colour contrast control in depth quality evaluation. The stimuli in Chapter 5 are 3D images comprised of practical scene design elements including shapes of models, gradient, objects texturing, lighting and shadows, instead of plain coloured squares in Chapter 4. The complex in the scenes increased the challenge to maintain the contrast ratios between foreground and background colours. For example, in the fountain scene, the foreground object, the fountain, contained different weight distributions of shadow and reflection on the object, and the shadow areas reflect lower brightness and saturation while the reflection areas mean higher brightness. Therefore, it is impossible to define a certain value for the object. The applied solution for this issue is to adopt value ranges in foreground and background colours to maintain certain ratio ranges of contrast. The brightest and darkest areas were measured to define the range of colour value then compare the range in the background to acquire the fitting contrast ratio. The approach of range measurement may bring the issue of lower data precision and reliability, which would bypass the aim of the experiment that was to measure the exact depth quality by different contrast adjustments. Moreover, the simultaneous contrast is an interaction with adjacent colours (Grandis 1986). The colour appearance in the 3D imagery stimuli was incorporated with numerous depth cues, such as gradient and shadow, which caught the difficulty to define the value of the adjacent colour. However, it still remains demanded to develop the proper approach to measure the contrast between complicated colour appearances. 148

163 6.4 Limitations The other major limitation in this study is the challenge to perform the device validation as described in Section 4.4. IEC standard is currently the only internationally verified guideline to perform the evaluation of the front projection device. However, the standard and procedure didn't describe conditions when applying to a stereoscopic projection system. The evaluation outcome would have the issue of insufficient validity and the result wouldn't be supportive to the experiment if this study performed the validation. In order to minimize the possible affection to the experiment outcome, this study utilised brand new DLP projectors and polarised filters to reduce potential fluctuations from aged apparatus, and moreover to have a more stable chromaticity constancy from DLP technology. However, without a device validation, the experiment may have higher uncertainty from the apparatus and the data obtained may still remain certain controversial. As the popularity of stereoscopic 3D projection is thriving, the demand of a verified standard for the specific stereoscopic projection system is worth to have more attention. 149

164 7.1 Research Summary CHAPTER 7 Conclusions 7.1 Research Summary In recent years the S3D filmmaking have grown to another peak since 1950s. Research into the depth perception of 3D scenes takes advantage of proper creation of depth cues, and in turn contributes important insights for stereoscopic depth enhancement. This thesis contributes to the recent trend in stereoscopic content design towards visual perception from psychophysical and depth quality view point. This work has focused mainly on stereoscopic depth perception of different coloured scenes and its applications in complicated 3D scenes. This thesis introduced the topic of human depth perception, stereoscopic vision, colour in depth perception and how they can be scientifically studied by using psychophysics and depth quality assessment. S3D filmmaking and current stereoscopic viewing technologies were reviewed to provide an overall perspective of S3D imaging development. Related works on the effect of colour on depth perception were discussed in a detailed overview. This work was based on the conclusion of previous works indicating that colour is influential on depth perception (e.g. Dresp and Guibal 2004) to investigate its effect under the polarised projection viewing condition. The data provided experimental evidence that colour can cause effect on stereoscopic depth perception and further verified that contrast is a more dominant depth cue than colour (e.g. O'Shea et al. 1993; Schor and Howarth 1986), even under the polarised viewing. The psychophysical experiments utilised to examine the thresholds of colour changes in saturation, hue and brightness between foreground and background colours indicated that saturation has it effect in S3D depth perception, but the studies claimed that 150

165 7.1 Research Summary the high saturated object appears nearer than the low saturated one (e.g. Luckiesh 1918; Egusa 1983; Bailey et al. 2006) is not applicable in S3D viewing. The contrast with its background is more determined than saturation. The similar result happened in hue. Viewers can notice depth change by manipulating in hue but there is no significant data to claim that long-wavelength colours appear nearer than short-wavelength colours (Dengler and Nitschke 1993) in S3D depth perception. The results in this study suggest that neither high saturation nor warm hue to be arranged in the foreground to influence the depth are major concerns when designing colour for depth enhancement. The similarity between the standard and reversed trials indicated that the foreground and background colour contrast is the more determined factor to influence S3D depth perception. The results from the brightness section show an interesting point of view about dominance in depth cues. Brightness section showed an overall poorer discrimination in depth perception under equal contrast ratios that applied in saturation and hue sections. The outcome can be explained by that brightness may cause the simultaneous contrast which has the effect of darkening the dark colour and lightening the light colour more (Grandis 1986). This attribute made brightness a more dominant depth cue and resulted a smaller value changes in brightness stimuli compared with saturation and hue stimuli in order to maintain the fixed contrast ratio to obtain comparable results. On the other hand, the results also indicated the phenomena that low brightness background enhances the nearness judgment for long-wavelength colours while high brightness background helps short-wavelength colours to stand out (Dengler and Nitschke 1993). The stimuli in the standard trial is a reversed arrangement to this phenomena and that resulted a very poor discrimination performance. The thesis continued to study the influence of colour on stereoscopic depth perception by depth quality assessment. Compare with psychophysical experiments using simple colour palettes in CGI environment, the assessment constructed detailed 3D scenes including a main foreground object and background objects to test colour s influence on stereoscopic 151

166 7.2 Possible Extensions and Future Works depth quality in industry-practicable scenes. The results indicate a corresponding conclusion to the psychophysical experiments. Under equivalent contrast ratios, saturation and hue showed similar depth quality performance, and only the arrangement of high brightness in foreground with low brightness in background showed a weaker stereoscopic depth quality than its counter colour design. The experiment results validated that it is insufficient to merely consider colour attributes when designing colour for a stereoscopic depth perception, but to consider the interaction between colours in foreground and background. Contrast is verified in this study to be the dominant factor in depth perception and worth to be the major consideration for depth enhancement. Brightness is a more effective value to manipulate the contrast, but it is suggested to be more careful in brightness adjustment as its influences in contrast may cause confusing to viewers depth judgment. Throughout this work this thesis has aimed to perform quality research with the intention of providing useful suggestions for colour design for stereoscopic depth enhancement under polarised projection viewing. Colour in human visual perception has made tremendous progress in the past decade, and this thesis may serve as a foundation and an inspiration for future work in colour design for S3D content, both in perception research and its application. 7.2 Possible Extensions and Future Works Various topics of interest related to the current work might be explored in future experiments. Some of the areas are quite specific and relate directly to the studies outlined in this thesis, and others are broader research areas and general questions of interest. A standard for stereoscopic polarised projection can be developed to give a complete device validation guideline for related works. The thriving of S3D 152

167 7.2 Possible Extensions and Future Works industry is triggering the demand of the research and development in the area. An investigation to build a verified standard to perform device validation would be in high demand. The stimuli in this study were manipulated by fixed colour contrast ratios and brightness showed an overall lower discrimination compared to saturation and hue. It would be interesting to test the stereoscopic depth perception by luminance contrast that is most determinant in symbols and text and highly associated with brightness. Luminance contrast can be seen as a contrast operated by brightness and a influential factor to dominant the depth perception. The results may be able to explain more details about the role of brightness in stereoscopic perception. Another specific question that might be explored is the possible elimination of binocular depth perception caught by inappropriate colour combination. Specifically, a weak contrast may have the potential to cause the elimination of actual stereoscopic depth that arranged in the scene. An example was described in Chapter 2 that as the advancing quality of the red, most of the time red card was moved by the participants further away in order to make it appear on the same plane with blue card (Luckiesh 1918). Luckiesh's experiment introduced that colour can affect viewers nearness judgement and adjust the advanced colour to have a physical distance to make it appear the same. It would be worth to quantify the moving distance of the red card and the relation with its colour attribute, and apply the similar procedure to stereoscopic depth distance in the scene with its contrast interacted from the foreground and background colours. The results may reveal a more specific relation between colour design and stereoscopic depth quality. The experiments and assessments in this thesis were carried out in the context of a computer-generated imagery environment. The results are applicable particularly for full CGI filmmaking. It would be interesting to do similar colour evaluation in real live environment with stereo camera rig to see whether there is a different result between CGI and real live environment. Throughout this thesis, colour is the only issue that has 153

168 7.3 Confirmation of the Thesis discussed in scene design for S3D perception. It would be interesting to explore more issues in scene design for S3D content production, such as lighting, object positioning and shapes, material perception and camera setting. It is expected that this thesis provide a solid foundation for future research and a source of inspiration for digital media artists getting started in the exciting field of stereoscopic content design. 7.3 Confirmation of the Thesis The original stated aims of this thesis was stated in section as: To validate this thesis, it is necessary to Investigate human depth perception and S3D content production and highlight the necessity of considering stereoscopy from outset in scene design, and then designate conceivable influence of colour perception on stereoscopic depth viewing. Build a polarised projection system that based on the current format of 3D cinema to view S3D stimuli. Produce S3D content for stereoscopic depth perception evaluation. Experimentally evaluate colour considering hue, saturation and brightness on stereoscopic depth perception through examining psychophysical thresholds. Experimentally assess stereoscopic depth quality considering hue, saturation and brightness in detailed 3D scenes based on ITU-R recommendations. Conclude the comparison of colour arrangements for S3D content productions. As discussed in the research summary, all of the above goals have been accomplished. It is possible to conclude that the original thesis has been shown to be well supported. Therefore the contributions to knowledge from this thesis can be concluded as following: 154

169 7.3 Confirmation of the Thesis A psychophysical evidence supported by a corresponding depth quality assessment result to show the influence of colour on stereoscopic depth perception which can be applied to colour design for stereoscopic depth enhancement. A conclusion based on statistical data to claim the common known advancement in depth perception from high saturated colour and warm hue is not applicable in stereoscopic depth viewing, instead, colour contrast between foreground and background plays a more dominant role in depth judgment. Brightness is a sensitive value that manipulated and intersected between multiple visual perception principles. Aspects have to be considered in brightness arrangement to achieve desired depth perception. A representation of polarised projection that can be implemented as a 3D cinematic viewing condition for future relative works. 155

170 156

171 Appendix A Appendix A Image Stimuli for the Psychophysical Evaluation in Chapter 4 Stimuli images utilised in Chapter 4 are shown in Tables A.1, A.2 and A.3. Table A.1 Image stimuli in the saturation section Contrast Ratios Saturation standard trial Saturation reversed trial 2.0 : : : : : : : 1 157

172 Appendix A Table A.2 Image stimuli in the saturation section Contrast Ratios Saturation standard trial Saturation reversed trial 2.0 : : : : : : : 1 158

173 Appendix A Table A.3 Image stimuli in the saturation section Contrast Ratios Saturation standard trial Saturation reversed trial 2.0 : : : : : : : 1 159

174 Appendix B Appendix B Instruction Sheets B.1 Instruction Sheet for Psychophysical evaluation Instruction Sheet for Psychophysical Evaluation Description In each trial, you will see two images side by side on the screen. Please focus on the small squares in images and judge which square (left or right) you perceive stronger stereo illusion. In other words, please choose which image perceived more distance between foreground square and background square. Press left or right keys on the keyboard Press the left arrow key if you perceive stronger on the left image; press right arrow key if you perceive stronger on the right image. Press left or right key only Control the pace on your own Each trial will display up to 30 seconds, however, you can press the left or right key once you make decision. 160

175 Appendix B B.2 Instruction Sheet for Depth Quality Assessment Instruction Sheet for Depth Quality Assessment Description In each trial, you will see two images side by side on the screen. The left-hand image is the reference image; the right-hand image is the test image. Please judge on the test image regard to the stereoscopic depth quality based on the reference image (score 5, fair) by a nine point scoring scale associated to terms bad, poor, fair, good and excellent. See the following figure that illustrates for the scoring scale. The stereoscopic depth refers to the distance between foreground objects (mushroom, fountain, bottle)and background objects. In other words, please score on the test image based on 161

176 Appendix B the image s strength of immersive viewing experience produced by the distance between foreground object and background objects. Press 1 to 9 keys on the keyboard Press on 1 to 9 keys only Control the pace on your own Each trial will display up to 30 seconds, however, you can press the left or right key once you make decision. 162