Abstract: Natural scene statistics has been widely used in blind image quality assessment,but most assessment methods are designed for gray images,and color space information is not properly used.We studied 5 color spaces (RGB,HSV,LAB,YCBCR,YIQ),and then used Gaussian distribution,logistic distribution,extreme distribution and T distribution to model the normalized coefficients,in order to find the best model for the color space.Then we used Gaussian model parameters as feature to classify the distorted images in LIVE database.We found out that the classify precision in some color space outperforms that in gray-scale statistics.

Key words: Image quality assessment,Natural scene statistics,Color space,Image processing