Reduced Reference Image Quality Assessment Based on Image Statistics in Pixel Domain

In this paper, we propose a new reduced reference image quality assessment (RR IQA) algorithm based on the image statistics. The image statistics is modeled in pixel domain, which is based on the gradient distribution of image. Compared with frequency domain coefficients, gradients are more easily calculated. The change of statistics in the gradient domain is measured to evaluate image distortion. To solve this problem, we fit the marginal distribution of image gradients to the integrated Weibull distribution locally. Then the estimated model parameters are extracted as the quality feature. We further propose a new RR IQA metric by quantifying the similarity between the original and the distorted quality features. Experimental results show that the proposed metric outperforms the well known RR IQA metric and has a comparable performance with the widely used full reference IQA metric Peak Signal to Noise Ratio (PSNR).