In encoding image sequence at a very low bit rate, the segmentation-based coding methods have been proposed. The block-based coding methods transmit the motion and the colour parameters of each block, while the segmentation-based coding methods, additionally, have to transmit shape parameter.
The shape information occupies a large portion of data. A lossless coding scheme for contour is not efficient, so lossy coding scheme is introduced.
The well-known method for lossy coding is polygonal approxmation. But, in this method, the subjective quality of the contour is degraded at a very low bit rate. So a new coding algorithm using polygonal approximation and VQ(vector quantization) has been presented. The subjective quality is improved with this method. But this lossy coding scheme still requires many bits.
In this paper, we describe a new lossy method for the contour of an image sequence coding using temporal redundancy. It is based on the motion compensation, the polygonal approximation and the VQ.
In order to exploit the high temporal correlation of the contour image sequence, the motion compensation is used for the corresponding label image sequence.
We improve the conventional VQ coding method by reducing the information of the polygon vertex using the grid transform method and reducing the area of the error region between the original and the approximated contour.
In this method, the contour is divided into the matched contour and the error contour. The matched contour is coded using the RLC and the error contour is coded using the polygonal approximation and the VQ.
Simulation results demonstrate outstanding performance for the coding of the contour image sequence. About 0.28bit/pel for the carphone sequence is required and 0.45bit/pel for the Miss America sequence.