Perceptual Video Quality Assessment Algorithm Based On Visual Attention Model

With the progress of video processing and communication technology,people's demand for video services are increasing rapidly,for example,digital TV,IPTV,video on demand,video conference,video monitoring and so on.Because of the huge size of original compression video signal,video compression is applied to different kinds of video service.In order to maximize compression ratio on the premise of ensuring that the quality can be accepted by people,the HVS are applied to algorithms of video compression.However,the distortion caused by compression algorithm and the distortion caused by transmission in error-prone network seriously reduced the perceptual video quality.Therefore,in order to provide better quality of experience of video services for the end users,plenty of scholars paid their attention to the problem of accurate evaluation of video quality.Firstly,we analyzed human's visual system.The principles of imaging in physiology and psychology are analyzed from the angle of nerve physiology and psychology respectively.Then the existing bottom-up and top-down visual attention models are studied.We made a comparison between subjective video quality evaluation methods and objective video quality assessment methods,which laid the groundwork for the following research.As the goal of our work is to provide reference for video encoding optimization,our research mainly focus on the all objective perceptual video quality evaluation method whose precision is good.Based on the above research,we firstly did research through modeling under the basis of five visual attention-skin characteristics,size of region,regional contrast features,motion characteristics,accident.Then we implemented nonlinear dynamic fusion on the obtained characteristics of the figure using two-dimensional gaussian filter.At last,we get the final visual attention model by considering the influence of gazing at a position.We use the AUC-ROC method to compare our algorithm with the existing salient map algorithm.The result show that our algorithm are more stable.Furthermore,we studied the contrast sensitivity of human eye.Starting from the static image,we considered the influence of motion and visual attention step by step and finally obtained the function of contrast sensitivity and the highest cut-off frequency that human is able to perceive.This provides ample preparation for the following perceptual video quality evaluation algorithm.Finally,based on the study of discrete wavelet transform,we chose the linear phase 9/7 biorthogonal wavelet basis to process the video signal and obtained the distortion index of space and time at the frame level.Then we deduced the final full video quality assessment algorithm under the application of asymmetric perception model.Comparison were made between our algorithm and several existing algorithms,which provided the theory basis and practice basis for future research on video encoding optimization.