Method and apparatus for a region-based approach to coding a sequence of video images

摘要

The present invention discloses a method and encoder for coding sequences of digital images for transmission or storage. Frames in the sequence are segmented into multiple regions of arbitrary shape each of which has a corresponding motion vector relative to a previous decoded frame. In a preferred embodiment, a hierarchical multi-resolution motion estimation and segmentation technique, which segments the frame into multiple blocks and which assigns a best motion vector to each block is used. Blocks having the same or similar motion vector are then merged to form the arbitrarily-shaped regions. The shape of each region is coded, and a decision is made to code additional image data of each region in one of three modes. In a first inter-frame mode, a motion vector associated with a region is encoded. In a second inter-frame mode, a prediction error for the region is also encoded. In an intra-frame mode, the intensity of each picture element in the region is encoded. A region interior coder with frequency domain region-zeroing and space domain region-enforcing operations is employed for effectively coding the interior image data of the arbitrarily-shaped regions. The region interior coder uses an iterative technique based on the theory of successive projection onto convex sets (POCS) to find the best values for a group of selected transform coefficients. The coded information, including the shape of the region, the choice of the mode, and the motion vector and/or the region's interior image data, may then be transmitted to a decoder where the image can be reconstructed.