Advanced video codec optimization techniques.

摘要

The rapid development of digital video transmission and storage has expedited the need for efficient video compression algorithms. Video coding standards have been developed for various applications. They recommend a general decoding methodology and syntax, but great flexibility is kept for the encoders. Improving coding efficiency and reducing the codec complexity are the two major challenges.;Decoder speedup is more practically important due to the much larger volume of decoders in the consumer electronic market. Technically, it is much more challenging to optimize the decoders under the constraint of being standard-compliant than to optimize the encoders. The deblocking process is identified as the most computationally expensive part for H.264 decoders. In this dissertation, techniques based on statistical analysis and efficient combination logics are proposed to achieve encoder/platform independent decoder speedup. The proposed hybrid scheme can reduce the deblocking computational load by more than a factor of seven times in comparison with the simplified H.264 reference software.;Realizing Video Cassette Recording (VCR) functionalities for digital compressed videos requires large computation and memory due to the motion-compensated prediction dependencies. This dissertation investigates the impact of decoding computational complexity and memory buffer when enabling VCR functionalities on the H.264 decoders. Two drift-free schemes called G-Group and Binary Reference Group of Pictures (GOP) Structures representing encoder side optimization are proposed. Theses schemes favor VCR functionalities with only 4.0%--7.6% bitrate increase while requiring much less decoder complexity and smaller buffer size relative to conventional GOP structures. The evaluation of the prediction schemes with comparison to the global optimal solutions is also presented.;Inter-frame dependencies are usually ignored in video encoder coding parameter selection. This gives a non-optimal solution and degrades the compression performance. A mathematical model to estimate the importance of each pixel, called PixelRank, is developed in this dissertation. Theoretical analysis on the parameters used for PixelRank calculation has also been given. The algorithm tracks the importance of each pixel and distributes the PixelRank scores. Therefore, the rate can be allocated more accurately, and better overall Rate-Distortion performance could be achieved.