Efficient motion and residual coding in motion compensated video coders

This dissertation studies problems within various areas of motion compensated (hybrid) video coding, including motion-compensated prediction, residual quantization and entropy coding. The overall goal of the work is to propose efficient cod ing algorithms that improve the rate-distortion (R-D) performance of hybrid video coders. An underlying theme of many of the proposed algorithms is that joint coding strategies are key to achieving overall coding gains. The major contributions are: (1) Motion-compensated prediction is formulated as a novel vector quantization paradigm called motion filter vector quantization (MFVQ). In MFVQ, the motion vector and the pixel-intensity interpolation filter are combined into a motion filter and the entire filter is vector quantized. The generalized Lloyd algorithm is modified to design unit gain and entropy constrained MFVQ codebooks. (2) An efficient entropy coding strategy, called the joint entropy wavelet residual coder (JEWRC), is proposed for coding a quantized motion compensated residual using the discrete wavelet transform. JEWRC jointly codes spatial significance in formation along with coefficient magnitude information using a very compact and efficient variable length code. JEWRC replaces the DCT and runlength entropy code in H.263+ and is optimized for low bit rates. (3) The construction of bit-rate and distortion predictors for low-complexity R-D optimized joint motion and residual coding is investigated. For computational efficiency, the predictors are implemented as lookup tables and treated as nonparametric estimators. For their design, a generalized scalar quantizer formulation is proposed and solved using a novel design algorithm that is based on dynamic programming.