Investigations of the display white point on the perceived image quality

Transcription

1 Investigations of the display white point on the perceived image quality Jun Jiang*, Farhad Moghareh Abed Munsell Color Science Laboratory, Rochester Institute of Technology, Rochester, U.S. ABSTRACT Image quality is generally estimated under a controlled environment. However, given the increasing number of images being viewed on observers' own displays, it becomes of interest to understand what might affect the perceived image quality in an uncontrolled environment. In the paper, an experiment conducted to learn the impact of the display white point on the perceived image quality by observers is discussed. Based on the experimental results, the perceived image quality by observers was invariant to the changes in the white point setting of the display. Keywords: white point, image quality, softcopy 1. INTRODUCTION Image quality assessment plays an important role in various image processing applications. 1 Perceived image quality can be evaluated with or without a reference. When a reference, usually an original, is available, image accuracy or color fidelity 2 is among the top priorities. Spectral imaging 3-5 is able to make the reproduction match with the original accurately in color under almost any light sources. On the other hand, when images are evaluated without a reference, preference rather than accuracy may be used more by observers to determine the image quality. 6 Given that more and more pictures are available online for personal use and public access, viewers usually do not have the original scene at their disposal. Images that appear more colorful tend to be preferred more even though subjects may realize these images look somewhat unnatural. 7 Nevertheless, provided more increment in colorfulness, the possibility of a straightforward preference for images with higher colorfulness is dismissed. 8 Preference rather than matching with the original overall becomes a more relevant contributing factor to the perceived image quality. Models have been built to predict the perceived (reference-free) image quality by considering it as a compromise between naturalness and usefulness. 8 Yet, such models may not be as successful with fine art reproductions, as paintings could be abstract or imaginative, thus bearing little resemblance with naturalness. In this paper, the influence of the display white point on the perceived image quality is investigated. Being one of the most important settings of the display, the white point can greatly affect the color reproduced on the display. Therefore, when images are evaluated with the original and accurate color reproduction has to be ensured, for example, a characterization of the display is a must at a fixed white point setting. However, if the originals were not present, and observers could fully adapt to the white point of the display, it would be of interest to understand whether a change in the display white point would still affect the perceived image quality by observers. In summary, the key contributions of this paper are two-fold: We conducted a psychophysics experiment to evaluate the impact of display white point on perceived image quality without the presence of the originals. We found the white point setting of the display statistically insignificant to the perceived image quality after analyzing the experimental data. *jxj1770@rit.edu Image Quality and System Performance IX, edited by Frans Gaykema, Peter D. Burns, Proc. of SPIE-IS&T Electronic Imaging, SPIE Vol. 8293, 82930O 2012 SPIE-IS&T CCC code: X/12/$18 doi: / SPIE-IS&T/ Vol O-1

2 2. EXPERIMENTAL A 30-inch Apple Cinema Display was used for the experiment with four white point settings selected. Their respective correlated color temperatures (CCTs) were 4500K, 5000K, 6500K and 9500K. 4500K and 9500K were the two ends of the white point allowed on a Macintosh computer. 5000K and 6500K were chosen because both are widely used as display white points. The white point was set by the display calibration tool provided by the operating system. No display characterization was conducted, as the change in color by simply altering the white point was great enough to learn observers preference on the perceived image quality. Figure 1. Test images Fourteen images varying widely in color and content were selected as shown in Figure 1. Among the test images, six were fine art reproductions, and the rest were photographs downloaded from the Internet ( or personal digital devices. Each test image was processed to have another two slightly tinted copies, appearing a little more bluish or yellowish than the original. By assuming the original image was of CIE D50 white point, the bluish copy was made by performing a chromatic adaptation transform from CIE D50 to a white point of 5100K. Similarly, the yellowish copy was obtained by transforming to a white point of 4900K from CIE D50. CAT029 was used for the chromatic adaptation transformation in the experiment as shown in Figure 2. For some test images, a shift in color was readily apparent, but the effect was not so obvious for other images. SPIE-IS&T/ Vol O-2

3 Figure 2. (a) The original image. (b) The bluish tinted copy. (c) The yellowish tinted copy. In Figure 3, the user interface composed of the three images (the original, bluish, and yellowish image) are shown. The background was made grey for observers to adapt to the current white point setting of the display. To ensure more complete adaptation to the display white point, observers were asked to look at the background for one minute before evaluating the image quality. The room lights were on during the experiment. The reason to create a yellowish and bluish tinted copy of each image is to learn whether observers would prefer a more yellowish image at a white point of higher CCT, as the original would now become bluish. Similarly, the bluish image would appear more neutral at a white point of lower CCT. However, if observers' adaptation state to the display white point were complete, a similar preference (the bluish, original or yellow image) should be observed independent of the display white point. During the experiment, observers were instructed to adjust the slider below each image on the interface (Figure 3) to indicate their preference. Once the adjustments were done, observers could click the Next button to proceed to the next image. Four observers were recruited to evaluate the fourteen test images at each white point setting, and therefore a total of sixteen observers participated in the experiment. All participants were from the Munsell Color Science Laboratory (MCSL) at the Rochester Institute of Technology (RIT). The observers were unaware of the white point setting of the display when doing the experiment in order not to bias the results. The order of the test images and the relative locations of the three copies (the bluish, original, and yellowish image) of each image on the interface were randomized before the experiment. To have observers fully adapt to the viewing environment (the white point of the display), the four white point settings were not randomized. Figure 3. User interface. The user interfaces were developed in Matlab using the Psychophysics Toolbox. The Analysis of variance (ANOVA10) was used to investigate what might contribute to the differences in the preferred images by observers. Several factors were identified in this experiment: wtpt (white point: 4500K, 5000K, 6500K and 9500K), triple (the SPIE-IS&T/ Vol O-3

4 original, yellowish, and bluish image), and image (14 test images). A 0.05 confidence level was used to distinguish significant factors from redundant ones. The statistical analysis was done in Minitab. 3. RESULTS The full model by including all three factors (wtpt, triple, image) and their interactions was tested. Wtpt and its interaction with triple and image were found to be insignificant. Therefore, the perceived image quality by observers was invariant to the changes in the display white point. On the other hand, the interaction between image and triple was found significant, and the interaction plot shown in Figure 4. Figure 4. Interaction between triple and image In Figure 4, the y-axis is the mean magnitude of the perceived image quality by observers normalized between 0 and 1. If the slider button in Figure 3 was not moved by observers, (the slider button was initially in the middle), the magnitude would be 0.5 on the y-axis. In Figure 4, the original images and the yellowish copies were perceived to be better by observers over the bluish ones overall. The Tukey method 10 was used to test the significance of the difference between the three levels (the original, yellowish, and bluish images) in triple by calculating the confidence intervals (CIs). The confidence intervals (type I error: 0.05) of the difference in the magnitude between the three treatment levels in triple are shown in Table 1. In Table 1, the CI of the difference in the perceived image quality between the original and bluish images excludes zero ([0.0102, ]). Therefore, observers preferred the original images to the bluish ones significantly in the experiment regardless of the image content and display white point. Similarly, the yellowish images were preferred to the bluish ones (CI: [0.0131, ]) significantly as well. However, the perceived image quality between the yellowish and original images was not significantly different, given that zero was included in the confidence interval of the difference (CI: [ , ]). Table 1. CIs between treatment levels in triple Lower CI Upper CI Original-bluish Yellowish-bluish Yellowish-original SPIE-IS&T/ Vol O-4

5 4. DISCUSSION An experiment was conducted to evaluate the perceived image quality of images shown on the display without the original. Given the absence of a reference, observers could not rely on image accuracy to evaluate the image quality. Color preference instead became a more likely candidate used by observers to rate the images. Based on the ANOVA analysis, wtpt (white point) together with its interactions with other terms were found to be insignificant for the test images in the experiment. It indicated that the images preferred by observers were not affected by the color of the display when images were evaluated based on preference. Also, when adapted to a certain white point, the observers would be likely to have the same preferred image independent of the white point of the display. Therefore, it became unnecessary to characterize a display when conducting an experiment to evaluate the image quality based on image preference. The result was in agreement with a previous experiment conducted online. 6 In addition, from the CIs calculated between the three treatment levels (bluish, original, and yellowish) in triple in Table 1, the images tinted slightly bluish were significantly lower in the ratings compared to the original and yellowish ones. Therefore, if a picture were taken without knowing the exact white balance, making it bluish would likely to reduce the perceived image quality rather than improving it. On the other hand, given that the interaction between image and triple is significant, the version (the bluish, original, or yellowish one) that was more preferred was image-dependent. For example, for Image 9 (Figure 5), the bluish copy was appreciated more by observers than the original or the yellowish one. In addition, the yellowish tinted images and the originals were found to be similar in the magnitude estimated by observers. Therefore, not much improvement should be expected on the image quality when images were made yellowish. Figure 5. The 9 th image. 5. CONCLUSION In this paper, we conducted an experiment to evaluate the perceived image quality on the display at different white point settings. By analyzing the data, we found display white point insignificant to the preference judgments by observers, despite that the white point could affect the color reproduction on the display dramatically. Therefore, we suggested that a characterized display was no longer needed when image quality was evaluated based on preference. An online testing was possible for no-reference image quality evaluation. SPIE-IS&T/ Vol O-5

6 REFERENCES [1] Z. Wang, A. C. Bovik, and L. Lu, Why is image quality assessment so difficult? Proc. IEEE Int. Conf. Acoust., Speech, and Signal Processing, pp , [2] D. A. Silverstein and J. Farrell, The relationship between image fidelity and image quality, Proc. of 3rd IEEE International Conference on Image Processing 1, pp , [3] L. A. Taplin and R. S. Berns, Practical spectral capture systems for museum imaging, Proc. of the 10 th Congress of the International Colour Association AIC Colour, pp , [4] L. T. Maloney, Evaluation of linear models of surface spectral reflectance with small numbers of parameters," J. Opt. Soc. Am. A, 3, pp , [5] L. T. Maloney and B. A. Wandell, Color constancy: a method for recovering surface spectral reflectance, J. Opt. Soc. Am. A 3, pp , [6] J. Jiang, F. Frey, and S. Farnand, Evaluating the perceived quality of soft-copy reproductions of fine art images with and without the original present, Proc. 19th Color and Imaging Conference, [7] M. D. Fairchild, [Color Appearance Models], Wiley-IS&T, [8] T. Janssen, Computational Image Quality, PhD thesis, IPO, Center for User-System Interaction, Eindhoven, the Netherlands., [9] N. Moroney, M. D. Fairchild, R. W. Hunt, C. Li, M. runnier Luo, and T. Newman, The CIECAM02 color appearance model, IS&T/SID Tenth Color Imaging Conference, pp , [10] A. Dean and D. Voss, [Design and Analysis of Experiments], Springer, SPIE-IS&T/ Vol O-6