A new technique for high quality low bit-rate image sequence compression

A system for coding image sequences is described. The system achieves a very high compression ratio so enabling video sequence transmission over low bit-rate channels. Image coding is obtained by motion estimation and compensation with a block-matching algorithm. Two techniques for coding the difference motion compensated image and the motion vector field, based on segmentation algorithms, are proposed. False motion vectors, common in conventional block-matching techniques, have been removed by means of thresholding and of a smoothing method. For smoothing, the vector field is modeled as a Gibbs/Markov random field. A mean compression ratio near to 100 and a PSNR always higher than 30 dB are obtained. Simulation results relative to different standard image sequences (namely Claire and CarPhone) are reported.<>