Efficient coding of segmentation maps for layered object representations

In object/region-based video coding, the side information is increased by the contour information. Thus, efficient lossy contour coding algorithms have to be developed. A class of very suitable motion segmentation algorithms provide a so-called layered representation of video sequences. In this paper an algorithm for segmentation map coding of layered object representations is proposed. It exploits the temporal connectivity of object masks by a transformation into an object domain. The contour coding is done in the object domain by a wavelet-based technique. Performance investigations show that very low bit-rates can be reached.