An efficient fast full search block matching algorithm with SSD criterion

Abstract

An efficient fast full-search block-matching algorithm is presented that uses the sum of squared difference (SSD) criterion. The efficiency of this algorithm mainly depends on the homogeneity regions of images in the video. If the blocks having small variance values are more in current frame then this algorithm takes far fewer computations for best block searching. The proposed algorithm is executed on a personal computer with 2.53-GHz CPU. Since the CPU takes fewer clock cycles for addition and subtraction operations than multiplication operation, this paper focuses on ways to decrease the number of multiplications to speed up the motion vector search. In addition to the temporal redundancy we also try to take advantage of the spatial redundancy for reducing the number of multiplications. In a simulation of motion estimation, it has proven that the proposed algorithm achieves exactly the same optimal results as the direct SSD full search, but with a processing speed of about 50 times higher.