Evaluating the Gaps in Color Constancy Algorithms

Transcription

1 Evaluating the Gaps in Color Constancy Algorithms 1 Irvanpreet kaur, 2 Rajdavinder Singh Boparai 1 CGC Gharuan, Mohali 2 Chandigarh University, Mohali Abstract Color constancy is a part of the visual perception system which permits people to recognize color in a variety of conditions, and to see some consistency in the color. Different color constancy algorithms are used like gray world, white patch, gray world 1st order derivative, gray world 2 nd order derivative. This work focuses on using the gray world 2 nd order derivative because this algorithm is based on 8 neighbors. The main objectives of this paper are to review the existing work on color constancy algorithms. The overall objective of this work is to find the limitations of the edge based color constancy using 2 nd order derivation and suggest suitable solution for the same. Keywords: Color constancy, Edge based hypothesis, Gray world, and Chromaticity neutralization. 1. INTRODUCTION Color constancy [1] - [11] is the capability to receive the color of objects in an image invariant to its illumination. The color of objects in an image is strongly dependent on the color of light source falling on an image. The different illuminations falling on an image alters or changes the appearance of images. Objects captured in different illuminations, loses its actual existence or message. Human vision has a natural capability to correct the effects of illumination falling on the objects of an image. The objects captured by digital camera under distinct light sources vary depending on the color of light source. The change in illumination depends on different factors like time of day, images captured in darkness and brightness. Color constancy [1] - [11] is the ability to recognize a relatively steady color for an object even under different illumination. Most of the methods are pixel-based, improving an image so that its statistics meet the assumptions such as the average intensity of the scene under achromatic light is world scene. Color Constancy is ability to recognize colors of objects in an image, invariant to the color of illumination. This ability is generally attributed to the Human Visual System, although exact information remains uncertain. An image consists of set illuminated objects. In general the light has a multifaceted spatial distribution, so that the light falling on one object in an image may differ from that falling on another. None-theless,a useful point of departure is to consider the case where the illumination is uniform across the scene, so that it may be characterized by its spectral power distribution, E(λ).This functions specifies how much power the illuminant contains at each wavelength. The illuminant reflects off objects to the eye, where it is collected and focused to form the retinal image. It is the image that is explicitly available for determining the composition of the scene. Figure 1 Image under different illuminations (adapted from [8]) Volume 2 Issue 4 April 2014 Page 1

2 However human eye has ability to display color constancy to large extent. Consider an example concerning only lightness, take a page of black print on white paper seen firstly under an indoor light and then under direct sunlight i.e. achromatic light. The intensity of the light reaching the eye from the white area of the page in indoor illumination is roughly equal to the intensity of the light reaching the eye from the black print in sunlight. (Kaiser and Boynton 1996, p.199) In spite of this rough equality, the page looks white under the indoor illumination and the print looks black under sunlight. 1.1 Gray World Gray-World [2] [7] is most familiar technique of color correction which considers the average reflection from surfaces of an image is white light. This assumption is held very well: in a real world image, it is usually true that there are a lot of different color variations. The variations in color are random and independent; the average would converge to the mean value, gray, by given an enough amount of samples. Color balancing algorithms can apply this assumption by forcing its images to have a common average gray value for it's R, G, and B components. In the case, an image is captured by a digital camera under a specific lighting environment; the effect of lighting cast can be removed by applying the gray world assumption on the image. As a result of estimate, the color of the image is much closer to the original scene. (a) (b) Figure 2 (a) Input Image (b) Result of gray world 1.2 White Patch White Patch method aims to place the objects that are exactly white, within the scene; by assuming the whites pixels are also the brightest (I = R+G+B). White Patch approach is typical of the Color Constancy adaptation, searching for the lightest patch to use as a white reference similar to how the human visual system does. In White Patch highest value in the image is white. White Patch algorithm is best suited for forest category. 1.3 Gray Edge Hypothesis In gray Edge 1 st order derivative 4-neighbouring pixels are considered. The first order derivative-based edge detection operator to detect image edges by calculating the image gradient values, such as Roberts operator, Sobel operator, Prewitt operator. (a) (b) Figure 3 (a) 4-neighbouring pixels (b) 8-neighbouring pixels Volume 2 Issue 4 April 2014 Page 2

3 The 8-neighbouring pixels are considered. Unlike 4-connected, in 8-connected more information for image correction is available. Gray Edge using 1 st order derivative does not proof to be efficient because each pixel considers its 4- neighbouring pixels. So, in this method not all information is available for color correction. (a) (b) Figure 4 (a) Input image (b) Gray world 2 nd order 1.4 Gamut Mapping The gamut mapping is most widely used color constancy method to achieve effective results. On average pixel values are used to estimate the reflectance on surface of an image. Gamut mapping is extended to integrate the numerical nature of images. It has been logically shown that the proposed gamut mapping structure is able to include any linear filter output. 2. LITERATURE SURVEY Brown et al. (2012) [1] presented color strength information which is used to enhance the color constancy. Color strength is basically a blend of both saturation and intensity to establish when hue in an image is reliable. There exist strong relationships between hue error and color strength. The main advantage of color strength model is that it can be used to calculate the reliability of color information there in a pixel. It has been shown that color strength model determines that the pixels with the lowest color strength. It will provide the maximum knowledge for color constancy for well-known color correction methods. Gijsenij et al. (2012) [2] has expanded available methods by applying color correction to every image area, instead of applying to the whole image. After local (patch-based) illuminant estimation, these estimates are combined into more robust estimations, and a local correction is applied based on a modified diagonal model. This methodology reduces the influence of two light sources simultaneously present in one scene. Quantitative and qualitative observations on spectral and real images have shown that the planned methodology reduces the influence of two different illuminations simultaneously there in one scene. If the chromatic difference between different light sources is above 1 degree then the methods based on the consistent illumination assumption is used. When the chromatic difference is below 1 degree then the scene can be supposed to contain one uniform illumination. Gijsenij et al. (2012) [3] has shown different edge categories (material, shadow and specular edges) to enhance the outcomes of the edge-based color correction methods. Edge-based methods utilize the image derivatives to evaluate the illuminants. The weights of edges are calculated using photometric edge categorization method with the hypothesis that such picture is illuminated by white light. It has been shown that weighted edge-based color constancy based on specular edges provides accurate illuminant estimates with the assumption of white light source. Shadow edges performs slightly worse than specular edges using weight map, but much better than material edges. Moreno et al. (2010) [4] shown color gradient with superior color constancy conservation properties. The method is based on the angular distance among pixel color representations in the RGB space. It is naturally intensity invariant, implying high strength against bright spots of specular reflections and dark regions of low intensity. An innovative chromatic gradient computation has been represented, which is chromatically coherent, conserves the color Constancy and provides better Color Edge detection. Volume 2 Issue 4 April 2014 Page 3

4 Cepeda-Negrete and Sanchez-Yanez (2012) [5] has addressed the results of two familiar color correction algorithms in combination with gamma correction for the image improvement reason. The method has been used, considered two well-known algorithms White-Patch Retinex (WPR) and Gray World (GW) to enhance the image quality to large extent. Gamma correction is mainly used for dark image enhancement and provides clear distinction of colors. Instead of using single algorithm for producing enhanced image, the combined use of a color constancy algorithm and then using the gamma correction provides a better result. It has been found that Gamma Correction after applying color constancy algorithm yields better outcome for dark images. Joze and S. Drew (2012) [6] shown the White Patch Gamut that is expanded to the Gamut Mapping Color Correction method, consisting the bright pixels of an image. New constraints based on the feasible White Patch Gamut are added to the ordinary gamut mapping constraints, new combined approach performs gamut mapping approaches as well as other famous color correction approaches. These new constraints are, great extension to the field and in fact can be more powerful than the real set of constraints themselves. Yu and Liao (2010) [7] shown a new color constancy method to improve the images in low-light conditions. The technique applies shades-of-gray algorithm to the active set of pixels across the image in order to reduce the disturbing effects of the image illuminant and highlights. The post-processing step of histogram clipping and gamma correction is also used in order to improve the global contrast and lightness. Results on large range of images show the effectiveness of the method In addition to this, method avoids halo artifacts or graying-out artifacts that are present in retinex approaches. Bianco and Schettini (2012) [8] has addressed how skin colors are used to guess the illuminant color. Face detector has been used to find faces in the scene and their corresponding skin colors to guess the illuminant color of the scene. The method has been tested on the number of algorithms and scenes in RAW format, using both a manual and a real face detector. The results obtain from method shows the effectiveness of approach. Ahn et al. (2013) [9] presented color correction technique by means of color correlation. The algorithms for color correction based on low-level stastics are generally used because of their simplicities, low computational complication, and suitable performances with sufficient parameters. The output of the method clearly shows that color correlation is important feature for color constancy as it provides better results than existing algorithm. The method proves out to be easy, quick, and effectively improving the color constancy. Lee et al. (2011) [10] have used well-known color correction methods for recovering and detecting the skin color. The standard deviation on all experimental image sets has been used for quantitative estimation. The outcomes of color constancy methods shows that color difference of skin can be reduced which are caused by color illuminants. Out of all methods Gray World proves to be better than others. Gray World method can be significant process for enhancing the performance of recovering and finding skin color. Akhavan and Moghaddam (2010) [11] addressed the problem of color constancy using neural network to estimate the illumination of light source and corrects the illumination by omitting the light source. The method is based on combining the output of four well-known color correction methods to obtain better results because it has been proved that combined methods work better on colored images. The method works in RGB color space because outputs of usual sensors and cameras are in RGB color space. Three different neural networks have been used for each color band. The combined method shows satisfactory results but it is possible to improve the results further. Bleier et al. (2011) [12] have presented methods of color constancy to estimate the non-uniform illumination. The proposed method is based on different color constancy methods to estimate the single illuminant. In proposed method, image is divided into super pixels and each color constancy method is applied to every super pixel. Then these estimates are combined into a single illuminant color. The error on these estimates was much greater than the errors under single illumination estimation method. As a result, the outcome made from non-uniform illumination provides better results as compared to single color illumination. 3. GAPS IN EARLIER WORK The survey has shown that still much improvement is required in the color constancy algorithms. It has been found that the most of the existing research has following limitations:- 1. Color normalization has been neglected to balance the color artifacts which will be presented in the image produced by the color constancy algorithms; as the modification is done in the image according to measured light source. 2. Effect of the Human visual system is also ignored. Because the modification done by the color constancy is based upon the measured light source; which can be efficient some time or may produce poor results in certain cases. So CLAHE is required to overcome this problem. Volume 2 Issue 4 April 2014 Page 4

5 3. Most of the existing research has taken the results on the available data sets; not much work is done by taking real time color source affected images. 4. ANALYTICAL SOLUTION The color constancy is a procedure that measures the influence of different light sources on a digital image. The image taken by digital camera depends on three aspects, the physical content of an image, the light source incident on an image, and the characteristics of the camera. The objects captured by digital camera under distinct light sources vary depending on the color of light source. The goal of the computational color constancy is to account for the effect of the illuminate.. The objects captured by digital camera under distinct light sources vary depending on the color of light source. Many traditional methods such as Grey-world method, Max RGB and learning-based method were used to measure the color constancy of digital images affected by light source. All these methods have an obvious disadvantage that the light source across the scene is spectrally uniform. This assumption is often violated as there may be more than one illumination on the scene. For example, inner scenes could be affected by both inner and outer light sources, each having different spectral power distributions. Improved Edge Based Color Constancy using CLAHE and Edge preserving using gradients methods is required to reduce the short comings of the earlier work in color constancy algorithms. The solution is seem to be justifiable and will have great impact on vision application because as edge based color constancy will reduce the impact of the light but it also reduces the sharpness of the image and also may result in poor brightness and also may lost its edges; so to remove this problem one will use an integrated effort of the edge based color constancy along with CLAHE and Edge preserving using gradients for efficient results. 5. CONCLUSION This work has reviewed some well-known color constancy algorithms. By conducting the review it has been shown that the 2 nd order gray edge algorithm provides quite better results than other algorithms. It is also shown that the most of the algorithms are point to point based but only gray edge hypothesis based algorithms are based on first order and second order derivations which makes edge based color constancy unique. It is also shown that the edge based color constancy can reduce the effect of multiple noises in efficient manner. In near future we will extend this work to use image filtering and image enhancement techniques to improve the performance of the available algorithms. However parallel programming may also be used to improve the speed of the available algorithms. References [1] Brown, Lisa, Ankur Datta, and Sharathchandra Pankanti. "Exploiting Color Strength to Improve Color Correction." Multimedia (ISM), 2012 IEEE International Symposium on. IEEE, [2] Gijsenij, Arjan, Rui Lu, and Theo Gevers. "Color constancy for multiple light sources." Image Processing, IEEE Transactions on 21, no. 2 (2012): [3] Gijsenij, Arjan, Theo Gevers, and Joost Van De Weijer. "Improving color constancy by photometric edge weighting." Pattern Analysis and Machine Intelligence, IEEE Transactions on 34.5 (2012): [4] Moreno, Ramón, Manuel Graña, and Alicia D'Anjou. "An image color gradient preserving color constancy." FUZZ-IEEE [5] Cepeda-Negrete, Jonathan, and Raul E. Sanchez-Yanez. "Combining color constancy and gamma correction for image enhancement." Electronics, Robotics and Automotive Mechanics Conference (CERMA), 2012 IEEE Ninth. IEEE, [6] Joze, Vaezi, Hamid Reza, and Mark S. Drew. "White patch gamut mapping colour constancy." Image Processing (ICIP), th IEEE International Conference on. IEEE, [7] Yu, Jing, and Qingmin Liao. "Color Constancy-Based Visibility Enhancement in Low-Light Conditions, 2010 International Conference on. IEEE, [8] Bianco, S., and R. Schettini. "Computational color constancy." Visual Information Processing (EUVIP), rd European Workshop on. IEEE, [9] Ahn, Hyunchan, Soobin Lee, and Hwang Soo Lee. "Improving color constancy by saturation weighting." Acoustics, Speech and Signal Processing (ICASSP), 2013 IEEE International Conference on. IEEE, [10] Lee, Woo-Ram, Dong-Guk Hwang, and Byoung-Min Jun. "Comparison of color constancy methods for skin color under colored illuminants." Digital Content, Multimedia Technology and its Applications (IDCTA), th International Conference on. IEEE, Volume 2 Issue 4 April 2014 Page 5

6 [11] Akhavan, Tara, and Mohsen Ebrahimi Moghaddam. "A new combining learning method for color constancy." Image Processing Theory Tools and Applications (IPTA), nd International Conference on. IEEE, [12] Bleier, Michael, et al. "Color constancy and non-uniform illumination: Can existing algorithms work?" Computer Vision Workshops (ICCV Workshops), 2011 IEEE International Conference on. IEEE, AUTHOR Irvanpreet Kaur has done her B.Tech from Amritsar College of Engineering and Technology, Amritsar, Punjab, INDIA in Computer Science and Engineering with 76 percentage. Now she is pursuing M.Tech in Computer Science from CGC, Gharuan, Mohali, Punjab, INDIA. Her area of interest is Digital Image Processing, Database Management, Computer Networks, Software Engineering, Computer Architecture. Rajdavinder Singh Boparai has done his B.Tech and M.Tech from Punjab Technical University, Jalandhar, Punjab, INDIA and Punjabi University Campus, Patiala, INDIA in Computer Science and Engineering.He has published his papers in various National and Internatonal Journals. His area of interest is Digital Image processing. Now he is doing job as Assistant Professor in Chandigarh University, Mohali, Punjab, INDIA Volume 2 Issue 4 April 2014 Page 6