VLSI architectures for accurate motion and disparity estimation using full-search block-matching and edge preserving non-linear smoothing

Flexible systolic architectures implementing the full-search block-matching algorithm for motion and disparity estimation are discussed. These architectures require sequential input but perform efficient parallel processing, and are easily cascadable for allowing processing with a variety of block sizes. Because motion and disparity estimation are ill-posed problems, the estimated motion/disparity field always contains some errors. The estimation accuracy can be improved by optimizing the motion/disparity vector field under the constraint that it must vary smoothly except at discontinuities. For this purpose, an Edge Preserving Non-Linear Smoothing Filter based on the Markov Random Field theory is discussed, and a VLSI architecture for its real-time implementation is presented