A hybrid model-based image coding system for very low bit-ratecoding

摘要

This paper proposes a two-stage global motion estimation methodnand a hybrid coding algorithm for the model-based coding. In the firstnstage, global motion is estimated by the feature-based algorithm. Thenestimated result is further refined by the gradient method. Thentwo-stage estimation algorithm is performed hierarchically to remove theninfluence of the local facial expression. In the estimation process, wenalso utilize the steerable pyramid to improve accuracy. The facialnexpression region is coded by the clip-and-paste method, and isnpredicted by a classified prototype coding technique. The prototypencoding technique can greatly improve the coding efficiency for thenfacial expression. The areas, which are difficult to be described byngeneric models, are encoded by the proposed hierarchical motionnsegmentation algorithm. We segment the image to different moving areasnwhich can be modeled by affine models in a split-and-merge manner. Thensegmented results with the estimated affine models can give goodnprediction for the unmodeled regions. Computer simulation results shownthat the proposed complete model-based coding scheme gives very goodnperformance in terms of peak signal-to-noise ratio and compression ratio