Research On Transcoding Among MPEG Standards And Implementation Of A Transcoding Plan

In all kinds of concrete application environments, diverse formats and standards are adopted in the compression, storage, and transmission of video data. After the digital video signal is compressed and encoded, and in many positions that applications are transparent to users, these kinds of compressed video streams need to be processed by transcoding. The compressed video stream transcoding, namely, is to process the compressed video stream from one standard or format into another which is more suitable for a special application environment. In this paper, the framework, format of the digital stream, process of the encoding and decoding, of the standard of the video compressed encoding of the series of MPEG. are introduced and summarized in detail at first, at the same time, similarities and differences between the standards of MPEG-1 and MPEG-2, as well as MPEG-2 and MPEG-4 are analyzed, which founds the base of the following research efforts on the transcoding. The compressed video stream transcoding can be sorted into two kinds:one of the kinds is to convert the encoded video stream from one compression format suitable for certain transmission environment to another suitable for another transmission environment, which is called homogeneous video transcoding, and the other one of the kinds is to convert the compressed video stream from one standard into another, which is called heterogeneous video transcoding. Aimed at the homogeneous video transcoding, the transcodings of the compressed video stream formats following the same encoding standard are researched in this paper, including the arbitrary size transcoding, bit-rate transcoding, and successive/interlaced scan conversion. The algorithm of the spatial domain arbitrary size transcoding when the conversion proportion is an arbitrary rational number L/M, as well as the reuse of motion vectors aimed at arbitrary size transcoders are put forward. The algorithm of reusing motion vectors can save very much operation, but the video quality is little different from that of transcoders which renew complete motive estimation, so is commonly adopted in transcoders. In addition, compared with the pixel-domain transcoding, the DCT domain transcoding can save the conversion pair of DCT/IDCT, which may decreases operation complexity, thereby, here the method of the DCT domain transcoding is discussed, consists of the algorithms of the DCT domain image down-sampling(mainly aimed at such position that the conversion proportion is relatively simple), DCT domain anti motion compensation and DCT domain motion compensation. Because the MPEG-4 owns lots of predominant performances, it will be widely applied in many fields such as the transmission, storage, and edit of the network video. But much of the present video stream is compressed and encoded according to the MPEG-2 standard, thus it is rather necessary to research and develop efficient transcoders of the compressed video stream from MPEG-2 to MPEG-4. Aimed at the heterogeneous video transcoding, and regarded the video stream transcoding among MPEG standards as the background, the transcoding from MPEG-2 to MPEG-4SP@ML (Simple Profile@Main Level) is researched in this paper, including the contents of the transcoding of the bit-rate, spatial resolution and temporal resolution. Firstly, similarities and differences among the algorithms of the standards of the MPEG-2 and MPEG-4 are given, some key problems of the transcoding design are analyzed, and from which five kinds of transcoding architecture of the pixel-domain and the DCT domain are derived, at the same time, performances of the corresponding five kinds of transcoders are compared and analyzed. Each of the transcoding architecture is a trade-off between operation complexity and video quality, and has its own strong and weak points respectively. Through the previous research and comparison of the transcoder architecture, the fast cascaded pixel-domain transcoder architecture which omitted the motion estimation is selected to experiment with the compressed video stream transcoder from MPEG-2 to MPEG-4. Based on this transcoder architecture, some techniques of transcoders are carefully researched such as the semantic-level conversion, encoding mode reusing, motion vector reusing, DCT coefficient processing, and so on. Thereinto, the key algorithms involve the motion vector reusing, frame mode conversion, mixed blocks processing, DCT coefficientmapping, etc. The motion vector reuse can save much operation, and ensure certain image quality through the modified algorithm of the motion vector. In course of spatial resolution transcoding, implementing down-sampling of the texture data by making use of the algorithm of the DCT domain spatial resolution reduction can overleap DCT/IDCT conversion, so as to the operation complexity is reduced more. A feasible video stream transcoding plan from MPEG-2 to MPEG-4 is put forward, which implements the conversion of the semantic architecture by software through the method of the partial decoding, converting MV and DCT coefficient multiplexing. The experimental results indicate that the plan can basically ensure image quality after transcoding, and the delay of conversion is little.