Image and Video Quality Assessment Based on the Similarity of Edge Projections

The goal of image or video quality assessment is to evaluate if a distorted image or video is of a good quality by quantifying the difference between the original and distorted images or videos. In this paper, to assess the visual quality of an arbitrary distorted image or a compressed video, visual features of the image or video are compared with those of the original image or video instead of direct comparison of two images or videos. As visual features, we use directional edge projections that are simply obtained by projecting vertical and horizontal edges detected by vertical and horizontal Sobel masks, respectively. Then, to assess the image or video quality, edge projections are compared using the similarity measures of one-dimensional histograms such as the histogram difference, histogram intersection, Kullback-Leibler divergence, χ-square test, and Bhattacharyya distance. Experimental results using LIVE data set and 140 video clips that are compressed with H.263 and H.264/AVC show the effectiveness of the proposed methods through the comparison with conventional algorithms such as the peak signal-to-noise ratio (PSNR), structural similarity, mean singular value decomposition, and edge PSNR (EPSNR) methods.