Standards conforming video coding optimization.

摘要

Digital video coding and transmission have become increasingly important due to the increasing popularity of multimedia applications. H.264/AVC is the state-of-the-art video coding standard which improves video coding efficiency significantly, with a cost of high computational complexity.; Standards allow researchers to develop optimization algorithms to optimize the video encoders for complexity reduction and for video quality improvement. Complexity reduction is important for low power, real-time applications such as wireless video. Video quality improvement is important for high quality applications such as video broadcasting, HD-DVD etc.; For complexity reduction, Motion Estimation (ME) and Mode Decision (MD) consume the largest portion of the computation power in the H.264 encoding. This dissertation investigates three fast ME strategies by MV merging and splitting for Variable-Size Block Motion Estimation (VSBME), to explores the correlation of MVs of overlapping blocks. In addition, an Inter-mode decision method is proposed by Fast Mode Exclusion (FME) and Dominant Mode Prediction (DMP). Experimental results show that the proposed fast ME algorithms reduce the search points to about 4% of those by the fast full-search, and that the proposed FME and DMP reduce the computation by an additional 25%, with negligible Rate-Distortion (R-D) degradation.; For video quality improvement, rate-control is a critical functional block. In this dissertation, a novel two-pass rate-control algorithm is proposed to achieve more consistent video quality. The Distortion-Quantization Parameter (D-Q) property is collected in the first pass. In the second pass, the appropriate D-Q curve is adaptively selected for different Macroblock (MB) types to predict the current MB's D-Q characteristic and determine its quantization parameter. The proposed method achieves better and more consistent video quality, without significant extra computation.; Due to the wide availability of MPEG-2 content and the better coding efficiency of the H.264/AVC, it is desired to transcode MPEG-2 to H.264/AVC for universal multimedia access. We propose a motion-vector reuse and coding-mode reuse process for MPEG-2 to H.264 main profile transcoding. Simulation results show that the proposed approach significantly reduces the computational complexity of the transcoder without sacrificing the rate-distortion performance.