| Nowadays, China has become the world's No. 1 in communications power scale. The construction of broadband network is continuously developing, and the quality of communication has been improved. And video communication is of no doubt becoming the hot topic. Nowadays, distribution of digital video was mainly based on MPEG-2．It is a former standard constituted 10 years ago and achieves limit of video compression efficiency.In this situation, ITU and ISO set up Joint Video Team including ISO/IEC MPEG and ITU-T VCEG in 2001．JVT's primary task was to develop H.26L draft to be a international standard. ITU's technology recommendation-H.264 about digital video compression was accepted and became official international standard in 2003．H.264 mainly include integer transform,spacial infra-prediction,1/4 pixel motive estimate,multiple reference frame,multiple size of block inter-Prediction,uniform entropy code table,Context-based Adaptive Variable Length Coding,Context-Adaptive Binary Arithmetic Coding and loopfilter, and so on.The aim of this paper is set as these: optimization and implementation on integer transform and quantization in H.264．The working platform is windows system using C language. In the first stage, arithmetic research refers to JM8．4, and software edition is based on x.264．Then it is transplanted into TI's TMS320 DM642 DSP system. DSP's edition is mixed program of C language and linear assembly language. Then H.264 CODEC that is transplanted and memory distribution are optimized further. Video capture is input to DSP through theⅥ, and code stream is output through VO. And socket program is used for network communication. H.264 utilize integer transform and quantization by multiplication and shift. All operations of integer transform use integer operation to insure zero mismatch reverse transform between encoder and decoder; core of transform is implemented only by addition and shift; also scalar multiplication can be integrated into quantizer to induce count of multiplication; quantization by multiplication and shift can avoid division and float operation to improve encoding efficiency.There is two size of integer transform and quantization in H.264-4×4 and 8×8．Choice of arithmetic is following several factor:--Don't adopt arithmetic that use more time.--Video format is D1 for which larger transform block can gain better compression efficiency. In image process, large transform block can utilize image's related information better and make power that is transformed more concentrated.--If the quality of image is equal, choose more efficient arithmetic under.After theoretical analysis and simulation, adaptive transform and quantization use more time. For quality of large size image, 8×8 transform's is better than 4×4's, and 8×8's efficiency is also better than 4×4's. Finally, 8×8 integer transform and quantization is adopted except intra 16×16 prediction (intra prediction including 16×16 and 8×8, inter prediction block size larger than 8×8) .On the C language edition, adjust structure of the function and structure, and use the common method of software optimization. Use Profile tool to analyze the performance efficiency of the software. For the part of transform and quantization is computation-intensive, so rewrite this part to linear assembly language and utilize specific instruction to minimize cycles on DSP edition. Then according to characteristic of memorizer and two levels' cache, optimize cache and put the code and data that are called frequently on chip. After that, achieve the requirement of improving efficiency. |
PDF Full Text Request