Research And Implementation Of Bitrate Reduction Transcoding From H.264 To H.263

Video compression is one of the core technologies of video processing. The International Standards Organization has launched out a number of video compression standards,such as H.263,H.264,MPEG-4,and so on. Each standard has its specific application field. Different terminals, which support different standard, may be in difficulty when they communications with each other. Video transcoding attracts more and more research interests to slove the problem.This paper focus on the issue about bitrate reduction video transcoding from H.264 to H.263. Because of its high complexity, cascade pixel-domain transcoder can't be applied in realtime project directly. In order to solve this problem, three aspect of key technologies related to high efficiency transcoder is focused in this paper, including the ratecontrol optimization of H.263,the fraction-pixel interpolation optimization of H.264 and efficient mode decision. Firstly, A method of preestimating the initial quantization parameter of I frame(Intra frame) is introduced into ratecontrol,and a method of borrowing bits from subsequent frames to smooth the bit streams is adopted,and these measures improve the performance of the PSNR (Peak Signal-to-Noise Ratio). Secondly, A new algorithm based on the idea of"on-demand interpolation"is proposed for interpolation, and the corresponding code is fully optimized using MMX(MultiMedia eXtension)instructions. Thirdly, a efficient mode decision algorithm is adopted in H.264-to-H.263 transcoding, which introduces motion vector re-estimate to avoid high calculated complexity, this method can greatly simplify the encoder of transcoding system and guarantee the acceptable degradation of PSNR at the same time. Finally, the three modules above are tested respectively. Comparison of the time and PSNR performance between the cascade transcoder and the efficient transcoder shows that the proposed efficient transcoder can promote the speed of transcoding by about 20%, as well as, reduce PSNR less than 1.4 dB,and subjective quality is good.