The Study Of Video Transcoding Techniques

With the rapid development of multimedia communication and video coding technology, many video coding standards have been established for various video applications, such as MPEG-2 for digital broadcasting and general high-quality video applications, H.263 for low bit-rate video communications, MPEG-4 for streaming video and interactive multimedia applications, etc. As digital video applications become more popular, there are increasing needs to dynamically convert the video between different formats. A video transcoder is a device that converts video from one compressed format into another compressed format with different bitrate, resolution or syntax. It is one of the key technologies for video delivery over heterogeneous networks and various terminals. The goal of transcoding research is to utilize the coding information and statistics of the input compressed video stream to facilitate the transcoding process and achieve the best possible quality-complexity tradeoff under given application constraints.In this thesis, we first review the existent video transcoding architecture and algorithms. Then, several techniques to improve transcoding performance are presented for transcoding between different standards, including macroblock mode selection, quantization parameters (QP) mapping, quality and rate control, etc.We first present two efficient intra macroblock prediction mode selection schemes. The first scheme is based on the texture classification method using discrete cosine transform (DCT) coefficients of original image. The second scheme is developed from statistical analysis of the distribution of best intra prediction modes. Compared with the straightforward, cascaded decoder-encoder implementation using the fast intra mode selection scheme in H.264 encoder, the proposed algorithms greatly reduced computational complexity with acceptable quality degradation.For inter macroblock mode decision, a direct mapping scheme is presented for real time transcoding applications and other two schemes: top-down macroblock partition mode selection and bottom-up macroblock partition mode selection are used. The top-down and bottom-up mode selection schemes were first introduced as fast mode decision algorithms for video encoders. By utilizing the coding information of the input compressed video stream, they are improved for accelerating the transcoding process. Compared with the cascaded decoder-encoder implementation using the full mode selection, the top-down and bottom-up mode selection algorithms achieves similar video quality while significantly reduce computational complexity. The direct mapping scheme further increases transcoding speed and keeps the same virtual picture quality.We then present a simple but effective rate-distortion model for video transcoder. It is used to design the quality control strategy of transcoder. Three main QP mapping rules are established for video transcoding applications with different quality and bitrate requirement. In the end, we proposed an adaptive joint quality-rate control scheme for various applications based on the parallel transcoding structure on multiprocessor system.