Genetic motion search algorithm for video compression

Abstract

A new approach to block-based motion estimation for video compression, called the genetic motion search (GMS) algorithm, is introduced. It makes use of a natural processing concept called genetic algorithm (GA). In contrast to existing fast algorithms, which rely on the assumption that the matching error decreases monotonically as the search point moves closer to the global optimum, the GMS algorithm is not fundamentally limited by this restriction. Experimental results demonstrate that GMS is more robust than other algorithms in locating the global optimum and is computationally simpler than the full search algorithm. GMS is also suitable for VLSI implementation because of its regularity and high architectural parallelism. >