Region-based coding of motion fields for low-bitrate video compression

Low-bitrate video compression requires a compact encoding of motion fields. We encode motion fields by segmenting, them into regions of homogeneous motion. Each region is formed by a collection of blocks. The segmentation is represented by the connectedness relationship between adjacent blocks. To reduce the amount of connectedness information, blocks may be of variable size, as determined by a quadtree structure. We use context-based binary arithmetic coding to further reduce the amount of bits required to code the segmentation. A segmentation algorithm based on rate-distortion optimization is presented. Simulations show that such a motion-field coding scheme achieves coding efficiency comparable to the state-of-the-art H.264 codec with modest improvements.