Extending block-matching algorithms for estimating multiple image motions

To estimate local motion from moving images, many algorithms have been proposed. Although the concept of the existing block-matching algorithm is very simple, it works well even under real noisy conditions, and hence it is often used in various practical application fields such as video processing. However, the existing block-matching method does not work well if a given analysis image block contains multiple image regions moving in different directions, and/or if boundaries of a given analysis image block and object boundaries do not coincide. To solve this problem, we extend the basic concept of the existing block-matching algorithm, and thus form two different algorithms for the analysis of multiple distinct motions in two different complex configurations of local motion. The one algorithm is based on the assumption that there may be multiple distinct image regions undergoing coherent motion within a given analysis image block, while the other algorithm is based on the assumption that multiple different moving image patterns appear superimposed within a given analysis image block. Through computer simulations, we demonstrate the potentiality of the extended block-matching algorithms