When a block matching must be performed between two frames, the matching criterion is usually evaluated using every pixel of the block. One of the fundamental hypothesis of block matching (BM) is that all the pixels in a block move with the same motion vector. According to this hypothesis, a good estimation of the motion could be obtained by using only a fraction of the pixels in a block for the computation of the correlation. It is evident anyway that if too few pixels are used to get this match, then there will be an inaccuracy in the motion estimates.

In this paper we study the impact of particular subsampling patterns that exploit the SIMD architectures. We point out the considerable computational advantage that exist for processors having a higher number of registers (e.g. IA64[1] compared to Pentium III[2]). Moreover we show that the quality impact of macroblock subsampling up to one fourth of the pixels, is very limited. In this way more operations could be dedicated to the search strategy that is the main responsible for the quality of the prediction, when a non exhaustive search is adopted.