Improvement of video coding efficiency for multimedia processing.

摘要

The recent trend in multimedia computing applications requires increasing bandwidth and computing power. As an integral part of multimedia computing, the efficiency of digital video processing is a very compelling issue.; Three novel methods based on a new solution framework are proposed in this thesis. These algorithms improve the efficiency of video encoding standards, such as MPEG and H.263, and can significantly speed up video encoding with negligible degradation in bit rate and picture quality. These techniques classify blocks in the video sequence to reduce computation. One of the algorithms takes advantage of a heuristically observed property in the DCT to improve encoder performance. The other two use an L1 norm-based metric, called the SAD, to direct encoder processing.; The novel methods are applied to a public-domain H.263 software encoder. Before doing so, well-known optimization techniques first improve the performance of the original public-domain code, resulting in a 230% speedup in execution time with no change to the encoded stream. Profiling of this optimized encoder reveals that further improvements in the motion estimator and DCT/quantizer promise to have the greatest impact on overall encoder performance. After evaluating existing algorithmic solutions which improve performance in those two areas, a subset of these solutions are incorporated into the optimized encoder to form the “basic encoder,” which is the basis of comparison for the newly proposed methods.; Results derived from the simulation of the three novel algorithms are promising. These techniques further speed up the encoder by up to an additional 287%, with mostly negligible changes in bit rate and picture quality. These techniques are very flexible, in that they are applicable to any block-based motion-compensated DCT coder, with no restriction on the type of motion estimation or DCT implementation.