In the circumstances we want to deal with, transmission channel is limited and global motion can happen by camera movement, and also there exist interest regions which are more important than background. So very low bit rate coding algorithm is required and processing of global motion must be considered. Also an interest region must be reconstructed with required quality after decoding because of its importance. But the existing methods such as H.261, modified H.261 are not suitable for such a situation because they do not compensate global motions, which needs large amount of transmission bits in motion information and degrades image quality. Also they can not reconstruct interest regions with high quality because they do not consider the fact that `interest regions are more important than background'. So a new coding scheme is proposed that describes a method for coding image sequences differentiating the number of bits between background and interest region.
In this scheme global motion parameters are initially estimated for two adjacent frames and global motions are compensated for previous frame, and next local motion vectors are calculated between global-motion-compensated previous frame and current frame. Then the blocks having similar velocities are merged into a same group by a hierarchical clustering scheme and quadtree segmentation is performed in order to represent the boundaries between each interest region and background well. After that, a quantization step size is allocated to a region according to its importance and residual data are quantized for the region using this step size, and also side bits such as motion information, frame start, and segmentation information are calculated. Finally quantization bits and side bits are sent to a transmission channel with preventing buffer overflow or underflow. Simulations show that the suggested algorithm performs well especially in the circumstances where background changes and interest region is small enough compared with that of background.
