Research On Fast Technique Of Video Compression Based On H.264

With the upgradation of information technology, video compression and transmission has become an important research subject in modern multimedia communication area. H.264 video encoding technology, as an efficient video compression scheme, palys an important role in digital product area. Meanwhile, problems like complicated realization process and low encoding speed still exist. The time-consuming models in H.264 encoder and decoder are investigated in this paper. The research goals include enhancing the speed of encoding and decoding, and reducing the complexity of algorithm realization. The research work of this dissertation contains five parts.Transformation and quantization are important steps for converting spatial residual signals into frequency signals, all-zero block detection is an efficient means to reduce the calculation times of transformation and quantization. In order ro raise the efficiency of all-zero-block detection and avoid the effect of the reconstructed encoding frames caused by error judgement, we analyzed the base matrices of integer transformation coefficients and found that all elements in the same sub-area of the base matrices were equal or inverse to each other in this paper,. Utilizing this distribution property, we can deal with the elements within the same sub-area together, and deduce a looser judgement criterion for all-zero block. It can detect a larger proportion of all-zero blocks in advance.The time proportion of motion estimation among the total encoding process is greater than 60%. In order to reduce the search points in the block matching process and relieve the calculation burden of motion estimation, we presented an adaptive fast search algorithm in this paper. The motion strength of current block is determined according to the MAD value at the start search point. For blocks with lower strength, we choose small diamond as search pattern, otherwise, hexagon is chosen as search pattern. In order to improve the accuracy of block matching, we search new three points within the hexagon according to the site of sub-optimal point on the edge of hexagon after the final hexagon search finished for long distance searches. Our method is more efficient compared with common fast search algorithms such as EPMVFAST, and at the same time the quality of the reconstructed encoding frames usually drops 0.02-0.04dB.In intra prediction, the prediction of current block is usually conducted according to the pixels from the neighbouring blocks in many different directions, and the direction with smallest prediction error is chosen finally. In order to reduce the number of searching intra modes, we advanced a fast mode selection algorithm combining the edge direction classification and edge vector projection together. For 4x4 luma blocks, the candidate prediction modes are determined by calculating the accumulated edge strength on many intra prediction directions for each pixel, and the mode with the smallest RD cost is chosen as the optimal mode. For 16x16 luma blocks, the prediction results of 4x4 blocks are utilized directly, and avoid the repeat statistics on edage vectors. Project the mode of each 4x4 block on the relevant direction of mode in Intral6x16. Then, calculate the accumulated edge strength in each direction and sort them with their value. In Intra4x4 prediction, our method only searches one mode in the least case, and searches three modes at most.In order to improve the efficiency of inter mode selection and reduce the search process of unnecessary modes, we proposed a fast mode selection algorithm based on the analysis of image texture and the classification of luma mean value. For current macroblock, we firstly determine the search range of modes according to the average abosolute luma difference between the current macroblock and the corresponding macroblock in the previous frame. Then, check the homogeneity of luma 16x16 block by calculating the average texture strength. If it is homogeneous, then search Inter16x16, Inter16x8 or Inter8x16 modes. Othrwise, judge the homogeneity between several blocks according to the luma mean value of each 8x8 block, and determine the candidate modes. For Inter8x8 mode, similar homogeneity examination should be conducted to decide if the smaller partition will be applied to prediction. Our approach could decide the optimal partition mode accurately, and the times of mode selection could be reduced effectively. In CAVLC decoding, in order to solve the problem of multiple matching in the decoding process of coeff_token codeword, we advanced a fast grouping algorithm based on the prefix features of codewords in this paper. We group the codewords from variable length code tables according to the number of '0' bits ahead of the first '1' bit. In the decoding process, we initially read a bitstream according to the maximum code length, and decide which group to search according to the number of '0' bits ahead of the first '1' bit. The mapping relationship between the number of consective '0' in codeword prefix and the number of information bits needed to match is established to solve the proplem of multiple matching. In the meantime, in order to realize fast matching of codeword within the group, we append all codewords and ensure that they have unified maximum length of information bits for each group. Moreover, the codewords within each group are sorted in asending order of information bits. In the decoding process, the position of the matching codeword in the group can be determined immediately according to the decimal number corresponding to the information bits read from the input bitstream. Only one match within the group is needed for our approach, and it enhances the decoding efficiency of codewords greatly.