Research On Video Quality Assessment Based On Image Structure And Pixel Information

Visual signal quality assessment, as an interdisciplinary research, involves many fields such as computational psychology, biophysics, signal processing, statistics, and even behavioristics. It has received more and more attention in recent years. With the arrival of the "wireless revolution", the number of handset users especially who use iPhone and iPad expanded rapidly. And the wireless network data traffic increased dramatically. Cisco Visual Networking Index (VNI:Visual Networking Index) report predicted that66percent of the2014global mobile data traffic will come from the video. This shows that the video business has become an inevitable trend of user demands. This also means that the user experience which is closely related to the visual signal quality evaluation presents unprecedented importance.Since the non-ideality of real-world environment, the final presented image or video contains a variety of distortion, which includes image/video capturing process distortion, processing process distortion, transmission process distortion, reconstruction process distortion and so on. Therefore, an image/video quality assessment which is suitable for a variety of distortion types is the most valuable and applicable method. This paper focuses on generic no-reference image/video quality assessment. It proposes the contribution in the following aspects:1)The methods of image quality assessment are classified in accordance with the scope of application, and a detailed study on common methods of no-reference image quality assessment is presented. Then based on this research generic no-reference image quality assessment’s design points is summarized.2) Based on structural information and pixel distortions, a novel no-reference image quality assessment method is proposed. It uses color information for image structural information extraction to get visual content structure map, and then evaluate the image quality by using the map to weight pixel distortions. Experimental results show that the new method can effectively process the artifacts of WN, JPEG, GBLUR, JPEG2000, Fast Fading and so on. And it accords closely with human subjective judgment.3) By introducing a visual attention model which is based spatial position, the proposed image quality assessment method is applied to video single frame firstly. Then the hysteresis effect of human visual system is used to limit the consecutive frame quality. Eventually considering the memory effect of human visual system, the revised single frame quality has been pooled into video quality. With this, the image quality assessment is complete extended to video quality assessment.