Edge Highlighted Structural Similarity For Image And Video Quality Assessment

Image and Video Quality Assessment is one of the most important research fields inmultimedia technology. Structural Similarity (SSIM) has drawn great attention in video andimage quality assessment recently due to its better consistency with human vision,efficiency and simplicity. Despite its popularity, it fails to measure the badly distorted orcross distorted images, it also fails to assess the fast-moving scene and interlacing. Thus,researching more reasonable quality estimation models should be significant in image/videoprocessing fields. In this paper, an improved algorithm called Edge Highlighted StructuralSimilarity for Video and Image Quality Assessment is proposed. The idea is based on thefact that the human vision is sensitive to edge information, and concerns much more on thedistorted edge information. The main achievements involve the following aspects:1. In chapter3, an improved algorithm called Edge Highlighted Structural Similarity(EH_SSIM) is proposed by valuing edges as the most important information in animage, and human eyes concern much more on the distorted edge information. In theproposed EH_SSIM, the edge regions are first divided from an image according to Otsumethod, then the edge regions which are obviously perceptual distorted are chosen byJust Notice Difference(JND) model and their distortion measures are highlighted. Thesimulation experiment database is from the LIVE Image database Ⅱ which containsfive different distortion images. Experimental results show that EH_SSIM is moreconsistent with HVS than SSIM，especially for distorted images with Blurred or WhiteNoise contaminated.2. Block artifacts and edge blurring are the important factors that greatly reduce videovisual quality. Therefore an improved algorithm called Edge and Block ArtifactHighlighted Structural Similarity for Video Quality Assessment (EBH_SSIM) isproposed in chapter4. Similar as in EH_SSIM algorithm, EBH_SSIM picks up edgeand block artifact regions according to Otsu method, and highlights their distortion measures respectively. In order to calculate the temporal quality of videos, framedifference is used to judge sequences’ movements, and weight of each frame is chosenbased on the degree of interest of human visual. The widely used VQEG FR-TV Phase-Ⅰdatebase is applied in the simulation experiment. Experimental results show thatEBH_SSIM is more accurate than SW-SSIM, MOVIE and MC-SSIM, but themonotonicity and convergence of EBH_SSIM’s scatter diagram should be improved inthe future.