Study On The Development Of A Uniform Color Space For Perception-based Image Processing

In image applications,typically images are transformed from RGB to a color space where lightness and chrominance channels are de-correlated so that they can be processed separately as does the human visual system.State of the art color spaces used to perform image processing include YCbCr,YUV,HSV,HSL,etc.These color spaces are computationally efficient but not perceptually uniform.A uniform color space should be able to predict perceptual color differences without any inter-dependence between its unique attributes:Lightness,Chroma,and Hue.There are many applications of uniform color spaces including color difference evaluation,color appearance,gamut mapping,lossy image compression,image segmentation,image quality evaluation,etc.There were four main goals of this study that include:(1)to investigate significance of using a perceptually uniform color space in image processing;(2)to compare performance of different uniform color spaces to predict perceptual data including high dynamic range and wide color gamut;(3)to propose a simple and efficient uniform color space to most accurately predict comprehensive and most reliable experimental data;(4)to verify the performance of the proposed uniform color space by applying indifferent image processing techniques for perception-based image processing.This work can be divided into two parts:(?)method for testing performance of different uniform color spaces,and(?)development of a new uniform color space.In the first part of the study,an image compression-based method was developed and used to compare the performance of ten different color spaces where five of them were RGB-based and rest five were XYZ-based.The RGB-based color spaces investigated here include sRGB,YIQ,YCbCr,YCoCg-R,and RDgDb.Whereas,XYZ-based color spaces tested here include CIE XYZ,CIELAB,CIELUV,IPT,and CAM02-UCS.These color spaces were incorporated into image compression algorithms including JPEG,JPEG 2000,and current modification of JPEG named CSI-JPEG.The performance of different color spaces was tested in terms of compression ratio,color difference,and visual quality.Three different tests were conducted to rank different color spaces and compression algorithms.In the first test,average color difference between the original and reconstructed image was fixed to a certain number and different compression ratios were computed.In the second test,compression ratio was fixed to a certain number and different values of average color difference were obtained.A third and final test was conducted to verify the simulation results by psychophysical experiment where results were presented in terms of Z-score.The simulation results showed that more uniform is the working color space,better compression performance can be achieved for same perceptual color difference.Similarly,for same compression performance,a more uniform color space gives better image quality.The results of the psychophysical experiment verified the simulation results.The results also showed that CIEDE2000 color difference best attributes with the visual image quality.The overall results showed that uniform color spaces performed better compared with RGB-based color spaces(sRGB,YIQ,YCbCr,YCOCg-R,and RDgDb)for their compression performance.Further,IPT and CAM02-UCS performed insignificantly better compared with CIELAB and CIELUV but significantly better compared with RGB-based color space.The second part of the study is to develop a uniform color space.In the new era of the display technology,it is possible to display wide-range of luminance and highly saturated colors.One of the challenges for color scientists is perceptual encoding of high dynamic range(HDR)and wide color gamut(WCG)color difference signals.A uniform color space is desired to predict perceptual color differences without any inter-dependence between its perceptual attributes.Further,the color space should have minimum computational cost for real time or quasi-real time image processing.Initially,state of the art uniform color spaces including CIELAB,CIELUV,CAM16-UCS,IPT,and ICTCP,were investigated using a comprehensive range of experimental data.To test performance of different spaces,comprehensive experimental data were collected.The experimental data sets include:(1)small color difference data(combined visual data(COMBVD)),large color difference data(Optical Society of America(OSA)and Munsell Color Order System),constant hue data(Hung&Berns,Ebner&Fairchild,Xiao et al.),and wide-range lightness data(lightness above and below diffuse white)with luminance range of 0 to above 7400 cd/m2.The results for prediction of visual data showed that all of the above mentioned color spaces have some pros and cons.A new perceptually uniform and hue linear color space was then developed.Initially,a structure similar to ICTCP was considered and was re-optimized for uniformity.The model was then extended by modeling two new equations to improve hue linearity and prediction of wide-range lightness,respectively.The results showed that none of the state of the art color spaces could predict the experimental data comprehensively and satisfactorily.The results also showed that proposed Jzazb,color space performed either best or second best to predict different sets of experimental data.Jzazbz gave second best(CAM16-UCS was the best)performance to predict small color difference data set and best for experimental data corresponding to large color differences.The proposed color space predicts the most accurate for MacAdam,and wide-range lightness data.Note that wide-range lightness and MacAdam data sets were used to test performance of test spaces for high dynamic range and wide gamut color difference signals,respectively,Jzazbz also accurately predicted the constant hue data sets similar to IPT and ICTCP,and outperformed CIELAB,CIELUV and CAM16-UCS to predict constant hue in blue region.The proposed color space was then tested by integrating into different image processing algorithms and performance was compared with state of the art color spaces.On basis of the overall test results,it was concluded that Jzazbz should be confidently used for perception-based image processing.Mainly,following two contributions were made in the science and technology in the current study.1.An image compression-based method was developed to test uniformity of different color spaces.2.A new uniform and hue linear color space was proposed for image applications including high dynamic range and wide color gamut image signals.