Advanced techniques for video codec optimization

Video coding and delivery over networks have experienced dramatic growth recently and attracted increasing research interests. In order to provide high-quality video delivery services, many technical issues must be taken into consideration including: how to achieve better coding efficiency to improve delivered video quality over limited bandwidth; how to reduce the computational complexity of current video coding standards for low-power wireless devices; how to efficiently adapt the pre-stored video bitstreams to formats more suitable for networks, etc. In this dissertation, we investigate various new techniques on standard-conforming video codec optimizations.; Global Motion Compensation is an important but computationally expensive coding tool in MPEG-4. We propose a new iterative approach for Global Motion Estimation (GME) from coarsely sampled motion vector fields. The proposed iterative method minimizes the fitting error between a set of input Motion Vectors (MV) and the MVs generated from the estimated motion model, using the Newton-Raphson method with outlier rejections. We investigate the application of the proposed methods to fast GME for MPEG-4 Advanced Simple Profile (ASP) encoding, MPEG-2 to MPEG-4 ASP transcoding, and MV-based error concealments. We also propose a non-iterative GME algorithm, which first fits multiple global motion models from groups of input MVs, then a robust estimate is formed by a histogram-based post-processing.; Multiple Reference Frame Motion Compensation (MRF-MC) is an important new feature in the state-of-the-art H.264 video coding standard. We propose a fast Multiple Reference Frame Motion Estimation (MRF-ME) algorithm that takes into account the correlation/continuity of motion vectors among different reference frames, which effectively reduces the computations of MRP-ME, and achieves similar coding gain compared to the approach in H.264 reference implementation.; New video coding standards enable a very large set of coding modes in order to adapt to local signal characteristics, and thus achieve higher coding efficiency. However, the encoder complexity for the mode decision module is greatly increased, especially when Rate-Distortion Optimization (RDO) is used. We propose several new algorithms to optimize the mode decision in the H.264 Extension encoder using a transform-domain approach. Simulation results show significant encoder complexity reduction with virtually no coding performance loss.