The Key Algorithms Research And Realization Of H.264 Video Compression Encoding System Based On DSP

The extensive application of digital video technology makes a lot of video compression standards raise continuously. Among the many standards, a new generation video compression standard H.264, comparing with previous standards, increases many of new technologies such as intra prediction, motion estimation with highprecision and multimode, 4×4 integer DCT transform and quantization, UVLC and so on. Because of these technologies, H.264 becomes the preferred standard in area of video compression research with its high coding efficiency and network adaptability. However, its perfect performance is based on the cost of increasing its complexity, which greatly limits its applications in real-time area. So researching fast coding algorithms of H.264 and designing a better real-time video encoding system of H.264 have a very important theoretical and practical value.DM64X chips producted by TI have been widely used in the field of video processing with their ultra-high frequency and parallel processing capability. DM642 is designed specially for video image processing and becomes an ideal platform for H.264 codec. In the paper, fast algorithms were studied and H.264 real-time video compression encoding system based on the DM642 has been designed.The design originates from one cooperative projects with some scientific research unit. It is the main research content of video processing part in project. Based on this application background, the system presents a very high real-time requirement. Therefore, it must take full account of this important indicator in the overall program of system.The thesis first introduced the basic theory of video compression and the development of international standards for video compression on the basis of the research of H.264 standard in detail. And then, it mainly introduced the major encoding principles of each module in H.264, and focused on motion estimation algorithm and intra prediction mode decision algorithm, which impact on encoding speed and quality highly. A new motion estimation algorithm based on multi-template switching according to prediction direction (PDTS) was presented; Intra prediction mode decision algorithm based on texture direction detection was improved. On the aspect of DSP, the thesis described the hardware platform of system. How to program on DSP was introduced. The code transplantation and optimization were focused on in order to make the system achieve the intended objectives. Finally, the system was tested on hardware platform and the result proved that the system is a good performance real-time video encoding system.