A fast block matching for SIMD processors using subsampling

A new block matching (BM) algorithm for motion estimation (ME) is introduced in this paper. This algorithm is formed of two different steps: the first step considers, inside the search window, a fewer number of pixels rather than the full search algorithm (FSA); the second step considers a sub-sampled macroblock (MB) for the operations that contribute to determine the mean absolute difference (MAD). The sub-sampling technique is meant to exploit as much as possible the single instruction multiple data (SIMD) technique currently available on the most common general purpose processors. The overall number of operations needed to compute the BM is considerably reduced. The simulations results of the proposed algorithm compared with the conventional methods are given. The proposed technique shows an improvement in terms of number of operations required and for the PSNR and MPQM behavior, for MPEG2 encoded sequences