Joint motion estimation/segmentation for object-based video coding

A video coding scheme is presented in which the coding is performed on individual moving objects. A Markov Random Field model is employed in finding the motion and boundaries of the objects. By guiding the object segmentation process with the spatial color information, meaningful objects representative of the real video scene are extracted. Furthermore, this enables a systematic treatment in handling the covered/uncovered regions, as well as the appearance/disappearance of moving objects. The rate for transmitting object motion and boundary is greatly reduced by use of temporal updating. The interior coding is performed by object-based subband decomposition. Simulations indicate promising results for low bitrate applications.