An efficient shape coding scheme and its codec design

Abstract

Shape coding is essential for video object based video compression such as MPEG-4. The shape mask specifies a video object's transparency and opaque characteristics during the video composition process. With shape information, we may perform shape-adaptive processing, e.g. shape adaptive DWT, to save both computation and communication overheads. We propose an efficient shape coding scheme called differential chain coding. Experimental results show that it can outperform all existing shape coding methods, including the well-known CAE method. In the decoding phase, we employ the concepts of change pixels and devise a new scheme to construct the binary alpha plane (shape mask) from the shape contour. Based on the proposed scheme, a novel VLSI architecture design for the shape decoder is developed. Synthesis and simulation results also indicate the design can process over 1,800 QCIF frames per second with very little hardware complexity.