Optimization Research Of Intra And Arithmetic Encoding For Video Based On High Speed DSP Platform

In recent years,the popularity of smart mobile devices has made video applications become more diversified and more high-definition,people are not satisfied with the visual experience brought by traditional standard video any more,there is more demand for ultra high resolution video,resulting in the rapid expansion of video data.Although the network transmission and storage technology is also in rapid development,still faces challenges of large bandwidth and low transmission delay.A more high-performance compression encoding technology is in urgent need for the new demand of video applications,HEVC that a new generation of video encoding technology emerge as the times require.Compared with the previous video encoding technology,HEVC achieved the qualitative leap,it can increase the compression efficiency exponentially,and does not cause the loss of subjective image quality.Because of the high computational complexity,its encoder is difficult to achieve the requirement of real-time applications.Therefore,this paper carries out the research,aims to improve the encoding speed of HEVC.First of all,the paper takes the advantage of powerful computing performance of multi-core DSP processor,and achieves the code transplant of HEVC encoder on C6678 which is selected as a hardware development platform.Meanwhile,before the transplantation,the source code was modified appropriately so that it can adapt to the DSP platform well.In addition,the method of linear assembly is applied to optimize the calculation of absolute transform error in intra prediction,that brings about the increasement of parallelism of instruction execution and the effective enhancement of the computing speed of the module.In view of the spatio-temporal correlation of video image,a intra prediction algorithm with fast encoding block classification decision-making is proposed.According to the statistical distribution of the depth-division similarity between current CTU and the neighboring CTUs,the algorithm decides the depth range of current CTU with the depth information of the neighboring CTUs,and the hit rate is 95%.Which minimizes the traversal depth range of CTU,skips the prediction computation of coding blocks that at some depth level,and terminates the partition of the coding blocks in advance.Meanwhile,with the depth information of temporal spatial neighboring coding blocks,the pre-judgement of the current coding block's depth is made,so as to avoids the encoding computation of coding blocks that at certain depth layers furtherly,reaches the goal of fast division of encoding block.For the CABAC entropy coding,a parallel CABAC strategy of conventional coding is proposed.Because the adjacent binary bins maybe has the same probability model in the process of encoding,the probability model of the latter bin can be obtained through the adaptive updation of the probability model of the previous bin at the point,which makes they obtain the probability models at the same time,so can achieve parallel encoding,and original serial encoding method is no longer used.In the implementation of the parallel coding algorithm,the parallel encoding frame is determined at first,and the computational formulas of encoding range,downside of the range and the probability model are deduces after that.Finally,via selecting the standard test video sequences,the performance of the optimization algorithms are tested under different quantization parameters.And the results are proved to be effective by comparing the results of encoding time and encoding performance before and after optimization.