Quantification of color motion picture quality considering human visual sensitivity

When a color video system with highly accurate color reproduction is designed, a good measure is required for evaluating the image quality including the color reproduction. There are some degradation sources in a total imaging system from image acquisition to image reproduction. Color random noise is one of major degradations which is often added in image capture phase. In this paper, we investigate the performance of the S-CIELAB (spatial CIELAB) and STCIELAB (spatiotemporal CIELAB) color difference as a quality measure in color motion picture degraded by random noise. In the experiment, we added a spatiotemporal noise to three kinds of still images and two kinds of motion pictures, performed the observer evaluation experiment, and found the noise level at which the noise is just noticeable. Color differences between images which correspond to just noticeable noise level by human observers were measured by CIELAB, S-CIELAB and ST-CIELAB space. In this paper, we report the performance of S-CIELAB and ST-CIELAB compared to CIELAB. Introduction When a color video system with highly accurate color reproduction is designed, a good measure for evaluating the image quality including the color reproduction is required. Though, for instance, MSE (mean square error) or PSNR (peaksignal to noise ratio) is a measure often used in the field of video compression, it does not necessarily correlate the subjective image evaluation by human observers [1]. Wang et al are proposing sophisticated methods taking advantages of known characteristics of the human visual system (HVS) [2]. However, those methods seem to be weighted on the luminance. As a conventional measure for color difference, the difference in the CIELAB color space is used. However, in evaluating color difference between images the pixelwise color difference still does not correlate the subjective impression. Spatial CIELAB (S-CIELAB) [3] is a measure for calculating the color difference between images which taking spatial sensitivity characteristics of the HVS into account and therefore has better correlation with subjective evaluation by human observers. Furthermore, a spatio-temporal CIELAB (STCIELAB) [4] has been proposed for color difference evaluation for color motion picture. In this paper, we investigated the performance of SCIELAB and ST-CIELAB for the evaluation of noise degradation of color motion picture. As a preliminary experiment, images degraded by spatio-temporal random noise are evaluated by the above mentioned color differences. The random noise is added to each of the opponent color components and the just noticeable level of noise is subjectively evaluated. Then the color differences between those noisy images and the original images were calculated and how such measures depend on image contents and opponent color components noise-added were investigated. Color difference in S-CIELAB and STCIELAB The processing flow of calculation the color difference in S-CIELAB or ST-CIELAB is shown in Fig. 1. Original R, G, B images and noise-added R, G, B images are respectively transformed to the opponent color components, A, T, D. Then, the spatial frequency filtering in S-CIELAB and the spatiotemporal frequency filtering in ST-CIELAB are performed. Those filtered images are transformed to X, Y, Z colorimetric values and then L*, a*, b* values. Finally color difference is calculated pixel-by-pixel and then the mean value is calculated.