Based On TMS320DM642 DSP MPEG-4 Video Encoder Realization And Optimzation

With the development of network and communication technology and the improvement of the people's living standard, we take more interested in multimedia applications and there has been tremendous progress in video compression and coding. They give rises to international standards for moving images, i.e., ITU-T H.261, H.263, MPEG-1, MPEG-2 and MPEG-4. The international Moving Picture Experts Group (MPEG) formally brought forward the new MPEG-4 Standard in 1999. MPEG-4 firstly introduced the concept of video object and contained object-based coding and model-based coding, and then supported content-based interactive and object-based scalable encoding. The establishment of MPEG-4 indicates that the compression of video is developed from the first generation of pixel-based coding to the second generation of content-based coding. The standard pays more attention to interaction and flexibility in multimedia communication compared with former international standard of video compression, it is different from those old standards. What MPEG-4 defined is not a concrete algorithm, but a format and a framework. Its coding system is open, so the researchers could freely add new effective algorithm modules with different applied requirements. As a result, it has played important role in interactive multimedia application such as digital television, real-time video monitoring, mobile multimedia communication, video stream and visual game in internet. There were already applications and products based on MPEG-4 video coding system by software, but the video embedded system attracted more attention and were popular. Because not only its cost is cheap and its features are stable, but also the operations are convenient.In spite of those advantages, the implementation of the embedded system were difficult to the computation of video data amount and complexity of algorithm of MPEG-4. Dedicated VLSI-based (Very Large Scale Integration) can be made for such computationally intensive schemes as either complete codec or core modules. These approaches are efficient since the design can be highly optimized for particular algorithms. Disadvantages of the approach are the large number of dedicated modules and their limited adaptability for an algorithm to be modified. Adaptation to a new coding scheme can result in a redesign. More flexible implementations are achieved by utilizing programmable DSP. Such an approach can shorten the time to market but also offers adaptability by reprogramming.With the development of VLSI techniques, the performance difference between DSP implementation and VLSI will become very small. TMS320DM642 is a new programmable high performance multimedia processing DSP, that is the product of the TI. The aim of my thesis is to implement and optimizate of MPEG-4 video encoder based on TMS320DM642 DSP for real time video communication. Firstly, we study video compressing principles and basic algorithms. Then we simply introduced various of international development and actuality of video compressing standards .We describe the content of the MPEG-4 in detail, which includes the characters of the MPEG-4 system architecture, media object description etc.. In the MPEG standards ,"Profile"are defined as a collection of encoding tools. Because the MPEG-4 standard contained a huge collection of different tools, each application can make use of only part of them. We considered of practical application and selected the simple profile as for the coding scheme. Secondly, we emphatically study the motion coding and texture coding in the MPEG-4 standard. Motion estimation and compensation plays an important role in video coding and it takes more than 47 percent of total encoding time. Motion search algorithm is the key part of the motion estimation, which would affect the feature of video compression result .We analyzed the typical search algorithms (FS: Full Search) based on block matching of motion estimation and diamond search, and found that the diamond search took priority of the cost of time by experiences, which can shorten more 3 times than the full search. We study deeply two kinds of quantization algorithms. By contrasting, we presented the H.263 quantization as for our video coding. But also we presented the predicted zero motion vectors algorithm in framework of the MPEG-4 encoder to reduce the count account in motion estimation and motion compensation. We study and realize rate contrlling algorithm. Thirdly, we introduce implementation of MPEG-4 video encoder and some optimization methods based on the hardware feature of DM642. At first, we study the hardware architecture of DM642, instruction system, memory architecture of DM642, pipeline and software explored flow. We referred for the XVID open source in C to implementation of MPEG-4 and transformed the source to the DSP.Then we took the hardware resource into account DCT transform and quantification, and made for use of parallel processing. At last, we optimized allocation in memory and put the often–called functions and data sections in in-chip memory. By profiling tools, rewrote the key cost module with liner assembly language for pipelined assembly instructions; adopted some complier options and so on. By utilizing those optimization approaches, our video encoder performance was greatly improved.We carried out the experiment to those methods and it was proved feasible. Finally, we use the standard testing sequences to value the performance of the video encoder. Our research results can be used to real-time video monitor requiring for more than 25 frames per second. We have not implemented many advanced ideas and profiles but simple profile and basic algorithm because the algorithm of MPEG-4 is not mature.With the development of the video study, the implemented function of MPEG-4 would be widely used in many fields especially in wireless video communication.