Multistage motion vector quantization for video coding

We present a novel multiresolution block-matching scheme-based on multistage motion vector quantization (MSMVQ) for videocoding. The MSMVQ performs the block matching on the coarse-to-fine basis where rough results of block matching are provided at the earlier stages of MSMVQ with fast computation, and accurate results are given at the later stages with additional arithmetic complexity. The multiresolution block matching process at each stage utilizes the results of its previous stages to further enhance the performance of MSMVQ. Simulation results show that the algorithm can be effectively applied to the applications where video decoders having different demands on the computational complexities and resolutions for block matching are desired to share the same video encoder.