Adaptive matching scan algorithm based on gradient magnitude for fast full search in motion estimation

Block-based full search algorithm of motion estimation has a major drawback of significant computational load in video coding. Due to this problem, extensive research in fast motion estimation algorithms has been carried out. To reduce an amount of significant computation of the full search algorithm, we propose a fast block-matching algorithm based on gradient magnitude of the reference block without any degradation of the predicted image. By using Taylor series expansion, we show that the block matching errors between reference block and candidate block are proportional to the gradient magnitude of matching block. With the derived result, we propose fast full search algorithm with adaptively determined scan direction in the block matching. Experimentally, our proposed algorithm is very efficient in terms of computational speedup and is fastest among all the conventional full search algorithms. Therefore, our algorithm is useful in VLSI implementation of a video encoder with a large search range.