Studies On Key Techniques In Fast Video Transcoding

With the development of the multimedia network technology and the widespread usage of the multimedia terminal, the needs of accessing the any multimedia at anytime and anywhere are impending. The multimedia applications should afford the universal median access (UMA) services in the heterogeneous environments. The video transcoding can process the pre-compressed bit-stream dynamically and convert one compressed bit-stream to another, which is the effective way to provide UMA. The dissertation studies some important problems about fast video transcoding. The main contributions are given as follows.1. Motion Vector Refinement: Bit rate adaptation transcoding adapt the network heterogeneity, which makes the reference frame changed. The motion estimation is needed. First, this section analyzes the error drifts in bit rate adaptation transcoding, including error drift caused by motion re-estimation and error drift of the pixel-domain transcoding compared with the compressed transcoding. A Motion Vector Refinement algorithm based on Rate Range (MVRRR) is proposed. The search window of the motion vector refinement relates with the rate control range. The algorithm sets the refinement search window according to the rate control range, which decreases the computing complexity and avoids the video quality degradation for the excess matching. The experimental results show that the Motion Vector Refinement algorithm based on Rate Range is effective.2. Motion vector composition in frame-skipping transcoding: Frame-skipping technique refers that the transcoder has no output frames for some input frames to guarantee the output bitstream consistent with the needs of the channels and clients to adapt the heterogeneous networks and clients. For the temporal prediction, frame-skipping make the reference frame dropped and the motion vector must be re-calculated. Forward dominant vector selection (FDVS) optimizes the motion vectors partly and improves the video quality. However, FDVS results in the motion vector out of range and intra-coded dominant-block problems. Based on the analysises of the influences of the problems on the different complex video, the Intra-Refresh FDVS (IR-FDVS) algorithm is put forward which adopts the intra refresh technique, solves the above problems and decreases the computing complexity. The experimental results show that the IR-FDVS is effective.3. Compressed domain fast transcoding architecture: Video transcoder always trades off the computing complexity and video quality. The complexity of the compressed domain...