An Invited Paper for the 1998 IEEE International Symposium on Circuits and Systems

Predictive Shape Coding Using Generic Polygon Approximation

Jong-il Kim and Brian L. Evans

Department of Electrical and Computer Engineering, Engineering Science Building, The University of Texas at Austin, Austin, TX 78712-1084 USA


We introduce an efficient, MPEG-4 compliant, predictive binary shape coding method that consists of (1) global motion estimation, (2) local motion estimation, (3) matched segment coding, and (4) residual segment coding. Global and local motion estimation use contour pel matching and knowledge of previously reconstructed contours. After motion compensation, we code the one-dimensional reference contour indices of the matched contour positions. The final step codes the mismatched contour segments using residual coding. We use a maximum shape distortion tolerance parameter (dmax), which is zero for lossless coding, for both motion estimation and residual coding. We apply the new shape coding method to MPEG-4 binary mask test sequences in QCIF and SIF formats for a wide range of dmax values. The key contribution of our method is in lossy shape coding in which the average coding gain is more than 100% over generalized differential chain coding.

The full paper is available in PDF and GNU-Compressed Postscript formats.

Last Updated 11/08/04.