Entropy-constrained motion vector quantization

The authors present a novel motion vector quantization (MVQ) technique for video coding. The algorithm, termed entropy-constrained MVQ (ECMVQ) algorithm, allows the codebook size and the rate for motion vector transmission to be independently prespecified before the VQ design. The computational complexity and the rate for video coding can therefore be effectively controlled. To utilize the rate available for block-matching, the ECMVQ algorithm employs an iterative design procedure similar to that of entropy-constrained VQ (ECVQ) design algorithm for codebook construction. During the design process, the motion vectors and the partitioning of training data are optimized one at a time until the convergence of the algorithm. Simulation results show that the ECMVQ technique outperforms the full-search block-matching algorithms and existing MVQ techniques. Therefore, the technique is effective for the applications where both low computational complexity and high rate-distortion performance are desired.