Based On Mpeg-2 Digital Video Compression Technology

This paper gives a deep study in every aspect for improving compression ratio of MPEG-2. A video encoder and decoder according with MPEG-2抯 standard are implemented. Experiments show that the encoder抯 compression ratio can reach 168:1 and PSNR can be 3 0.78db in the bifrate of 5Mbit/s for the studio video sequence. Above all this paper discusses the necessary of video encoding, reviews the MPEG series and some basic point such as the data struct of video coding as a whole. Then the use of traditional compression methods into the MPEG-2 standard in detail, for example DCT transform, motion estimation and compensation, uniform quantization, run length coding and variable length coding, etc are discussed. In addition, the paper illustrate the improved aspects of MPEG-2 ?s over MPEG- 1, such as two AC coefficient抯 scanning methods, DC coefficient抯 precision and their effects on encoding. The importance of quantization matrix has been testified in the experiment. Bit-rate control for encoder and adaptive quantization is another emphases. Finally this paper gives some useful experiment results and the analyses of these results. The contributions of this paper are: it analyses some methods of motion estimation particularly, whose differences are showed by pluging into the encoder. Secondly, several prediction modes, such as frame/field prediction and the adaptive chosen of frame/field DCT transform to increase the efficiency of encoder are discussed. Fine research on two particular prediction mode in MPEG-2, dual prediction and 16 X 8 prediction mode, and a implementation of algorithm which adaptively chooses the prediction modes to improve the encoder抯 quality are developed. The experiments show a better performance, which not only increases the compression ratio but also enhanced the PSNR, than that of using a single prediction mode. The paper analyses the bit-rate control method and theory to ensure the output bit-streams have a constant bitrate through adjusting the Mquant value continuously. The paper can be referenced by VLSL implementation of MPEG-2 and complementary to encoding of MPEG-2 standard.