Video Quality Assessment Algorithm Based On HVS Properties

Video distortion will be caused by various noises and encodingprocess in transmission. Effective evaluation of video quality plays an important rolein processing applications. Human is the ultimate recipients of video, video qualityassessment (VQA) is the construction of the model results to achieve consistent withhuman senses. Therefore, the model of video quality assessment based on humanvisual system properties is the most idea way to evaluate. In this paper, we establishthe model of video quality assessment based on the human visual system (HVS)properties.First, this paper focuses on two important factors, masking effect in temporal ofhuman visual system properties and temporal distortion fluctuations. In the existingvideo quality evaluation studies, few algorithms take fluctuations of temporaldistortion into account to describe subjective perception quality of video.Improvements are as follows:(1) Motion is an important feature of video. It can reflect the temporalcharacteristics of video. This section analyzes the impact of motion in assessing videoquality. In the meantime, according to the motion characteristics, we measure thedistortion of adjacent frames.(2) According to the fluctuations of distortions in the temporal pipeline, wemeasure temporal perception distortion of video.(3) Modulation of spatial foveal vision and visual staying phenomenon act ontemporal distortions. Through the analysis of human visual system properties, weregulate the entire temporal model of perceptual distortion algorithms.Next, as the video quality assessment includes not only temporal perceptualdistortion. In order to further study the video quality assessment algorithms, weconsider adding the human eye tracking mechanisms. The research mainly includes asignificant distortion area and moving area.(1) Visual interested of the viewer varies in the different areas. People focus onthe distortion areas in video. We found that distributions of the temporal distortion and temporal fluctuations are the most important feature affecting visual interest.According to the quantitative measure of temporal distortion and temporal fluctuation,this article adopts an adaptive threshold to calibrate the possible points of interestingpixels. The spatial connectivity analysis is required for these selected pixels. Finally,in accordance with a clustering algorithm determine the final area.(2) The motion area detection is another part of the research of visual attention.We study and found that the amplitude of video frames and phase information occupyan important position for the detection of significant motion region. Video-basedanalysis of the spectral characteristics of adjacent frames, spectral differencecharacteristics of test frame and its neighboring are incorporated into video motionarea detecting research. Ultimately, target detection algorithm based on motionregions is derived.Finally, we have fused temporal perceptual distortion, saliency detection ofdistortion areas and motion areas in the final. According to the distortion regionsaliency and motion region saliency adjusting the temporal distortion, we establish themodel of video perceptual quality assessment, then testing in the LIVE database.Experimental results show that the algorithm and subjective evaluation scores arewith good consistency. These also verify the feasibility of the proposed method.