Corner-based Structural Similarity For Image Quality Assessment

With the development of high-tech network, a large number of images are transmittedin the transmission process. In the process of transmission, the compression and othertreatments of the image will lead to a variety of distortion. Nowadays, the internationalacknowledged and the most reliable approach of evaluating an image or video quality isMOS, but subjective methods are time-consuming and expensive, and they can’t beembedded into real application. Nevertheless, traditional objective methods such as peaksignal to noise ration (PSNR) and mean square error (MSE) have low correlation with theperceptual visual quality. So, it is necessary to develop new objective methods, which cancorrespond better to subjective feelings.Finally combined with the human visual system characteristic and the goodperformance of the corner, this thesis improved structural similarity algorithm from twoaspects：(1) weighted index improvement;(2) increase the corner comparison function onthe basis of weighted improvement. Weighted improved algorithm taking into account theblurred image, the human eye is more sensitive to contrast and structure than luminance.Consequently, we should use a more reasonable weighted function; Corner is a veryimportant local characteristic of image, which is a point of intensity change in the image. Itdetermines the shape of the target image, so it is very important role for the image,especially the blurred image to understand and analyze. Consequently, this thesis increasesthe corner comparison function on the basis of weighted improvement. The experimentalresults illustrate that the proposed metric approximates the mean of score (MOS) morethan the metric of the SSIM.