Visual Perception Based Video Quality Evaluation

Recent years, with the rapid development of broadband digital networks, the advent of Large Scale Integrated Circuit (LSIC) technology and the high-speed multimedia special chip, all kinds of multimedia technologies have been widely used in life. The digital image and video have widely seeped into all aspects of life, such as video surveillance, video conference and video on demand. However, the quality of images and videos will inevitably suffer distortions due to the acquisition system, transmission channels and processing algorithms. The images and videos'quality degradation can reflect the performance of the processing systems. Thus, it has great practical significance to evaluate the quality of images and videos.Human are the final receivers of information from videos, so the human eyes are the most effective instruments to assess the quality of images or videos. It could improve the performance of algorithms if we take the characters of the human visual perception into account. Visual attention (VA) is one of the most essential visual phenomena of the Human Visual System (HVS). It shows that the salient regions and objects in visual field are highly focused by human eyes and only partial information is processed at any time. The thesis proposes the method which uses the salient model KR and structural similarity algorithm to get the quality of videos. Experimental results prove that the correlation of the proposed method with subjective valuation has been improved a lot.A reduced-reference video quality assessment approach has been proposed. It is based on texture masking, video motion and scene analysis and model the Masking Function, which is an important feature of HVS. Artifacts are not easily seen in high textured region. The experimental performance of this algorithm is better than that of PSNR and VQM.The thesis explores NR Quality Assessment of MPEG-2 Coded videos. The blocking effect is the most obvious distortion in such block-based coding. The features of Zero-crossing are extracted from distorted videos to assess blocking. Then the obtained values are corrected by Correction factor. Experimental results prove that the final method achieves a better correlation with subjective evaluation than PSNR.