Research On High-Performance Video Compression Technologies

In recent years, video compression has become one of the most popular research technologies in multimedia research domains, which can be divided into two categories: one is high-performance video compression technology, i.e., given a certain bitrate to get compressed high-quality video data; The other is high-efficiency video compression technology, i.e., fast video compression encoding and decoding technology. With the rapid progress of high-performance video compression technology, it has been developed to be the integration of techniques, which includes the selection strategies for block-matching criteria, intra prediction coding techniques, motion estimation techniques, multi-frame reference video encoding techniques, and so on. But there are many problems for us to solve to further improve video compression performance, such as mismatching problem for matching criteria, temporal and spacial redundancy elimination of imperfections, compression efficiency vs compression performance, and others.In the dissertation, the key techniques of high-performance video compression coding are studied.First, in order to solve the mismatching problem for matching criteria, we propose a new block-matching criterion of NSAD (New SAD) and the corresponding block-matching algorithm. After anylising the weak coverage of changes in the shadow or the arithmetical changes of the brightness in video sequences, we present the linear expression of two data blocks with linear relationship, remove the constant weight, and then compute the value of SAD for the blocks so as to solve the mismatching problem. Compared with the criterion of SAD, the proposed block-matching criterion and corresponding algorithm could get a better video compression performance and improve the coding efficiency while maintaining negligible increase of coding complexity.Second, in order to solve the problems in the research on intra prediction modes of video coding techniques, the research work on intra prediction modes is carried out from two aspects in this dissertation. In order to select the actual intra-prediction mode quickly, a Star algorithm is proposed. Given an intra block, there will always be a best trend for variety of pixels' values in the block. By calculating the trend of variety of each pixel's value along the Star direction, we can get the best candidate mode so as to achieve the rapid intra prediction mode selection purpose. The thesis also proposes an Intra-Prediction Mode Based on Symmetry for H.264 in order to eliminate the redundancy between adjacent pixels in different blocks. Through searching the symmetry blocks for the current block in the encoded and reconstructed, we get one prediction block, and then put it into the intra prediction mode selection process to achieve the elimination of data correlation between adjacent pixels.Third, in order to eliminate the temporal redundancy brought by three-dimentional translational motion in the video sequences, we propose a global translational motion estimation (GTME) algorithm. The traditional motion estimation techniques could not describe the actual translational directions of moving objects in video sequences or the relative movement of the three-dimensional characteristics of camera, which is called zoom feature. With an idea to do the motion estimation block matching in the third-translational direction, perpendicular to the image plane, the GTME vector and algorithm are proposed, which can eliminate temporal redundancy more accurately, through interpolation techniques to implement the third-dimension translational direction of block matching, so as to achieve a better video compression performance.Finally, in order to reduce the computational complexity of multi-frame reference selection algorithm, we have studied the prarllel techniques about the multi-frame reference selection algorithm, and propose a model of video coding based on Multi-Agent System and three kinds of parallel pipelinings to work with the mechanism of cooperation, which are Forward Parallel Pipelining, Backward Parallel Pipelining and Whole Parallel Pipelining. The inter-frame encoding process could be executed in a parallel way. In order to make the multi-frame reference selection algorithm be parallelized, we put the entire video-coding task into a Multi-Agent distributed system and make the video compression coding be processed in a parallel way by FrameAgent's mechanism of cooperation, so as to improve video coding efficiency while maintaining high video compression performance.