Video Quality Assessment Based On Features For Semantic Task And Human Material Perception

With the development of Internet,multimedia communication between people becomes more and more frequent and important,so in the communication if the guarantee can get relatively good quality of image and video has become a valuable research topic.In the field of multimedia technology,how to measure the quality of multimedia digital signal and optimize its rate and quality is the key technology.In this paper,we focus on the development of a video quality assessment method based on the characteristics of human perception and semantic tasks.A meaningful and novel perceptual video quality metric should be based on features used for semantic task and human material perception for video quality assessment.Based on the neuroscientific and psychophysical evidence,from preprocessed video information,texton,shape,contour and color can be important mid-level cues for semantic task.And based on Human Material Perception studied,bilateral filters were used to separate structural information from the texture information for material recognition.Finally several estimators are gathered together with different weights.All of these are according to the spatio-temporal tubes scheme with block-wise motion estimation.The basic principle and technique in the evaluation of video quality were studied and the main study,some feature extraction is consistent with human visual characteristics based on the principle of combining free energy distribution combination feature weights for final measurement experiment.The thesis is organized as follows:first introduces the research status of multimedia quality evaluation at home and abroad;then introduces the process and method of performance measurement and the subjective and objective quality evaluation;then reviews the basic theory of excellent channel analysis and feature extraction;finally,the latest video database of the University of Southern California,the quality evaluation of our proposed system in the experiment,and its performance is compared to existing methods,the specific contents are as follows:In our proposed quality evaluation system,we can operate the subsystem of thesource video and the distorted video almost simultaneously according to the design idea of the simplified reference system.In each of the two subsystems,the image of each successive frame of the video will be separated by a block,and then the local motion coherence vector can be combined into a three-dimensional pipeline in time space.Then we will do channel decomposition and feature extraction on the corresponding pipeline or container.Then the diversity of the corresponding channels in the source video and the distorted video will be calculated and aggregated by the pooling strategy in time space.Finally,the nonlinear mapping model will make our objective video quality evaluation more close to the subjective experimental value.