Application Of Human Vision System Theory In Video Coding

Following the developing path of multi-media technology, digital video is becoming more and more important in human life. The electronic technology, computer system and telecommunication engineering of modern society provides a good platform for the applications of digital video, and also some restricting factors. The most one is that there will be no enough storage space and bandwidth to store and transmit the tremendous video data. To research this problem, this paper presents some methods to reduce the redundant information based on the human vision system theory.The main work and innovation is as follows:1 A method, based on the contrast sensitivity function, is presented to guide the quantization process of residual frames. First, we identify the information which is relative more interesting to human eyes by compute the contrast sensitivity function using the motion vectors in the video coding. Than some information of residual frames is discarded by judge the function values. This method is applied in the encoder of X264, it saves about 2%~5% bits than the result of the original encoder.2 A method, based on the luminance masking theory, is presented to guide the quantization process of residual frames. According to the luminance value of the frames and macro-blocks, the quantization parameters is well adjusted to improve the quality of the video. This method is applied in the encoder of X264, it does not increasing much bits than the result of the original encoder.3 A Double Stimulus Continuous Quality Scale II test is done to test the quality of the video. It shows that the videos reformed by our methods save about 2%~5% bits than the result of the original encoder while the quality is not degraded.4 A method to compute the motion vectors between the current frames and neighbor frames by using the motion vectors between the current frames and reference frames and the information in Temporal and View axis is presented. The method in 1 is transplanted based on these motion vector (between the current frames and neighbor frames). It does not increase bits than the result of the original encoder while the quality is not degraded.