A fast matching criterion for VLSI implementation of block-based motion estimation

The main bottleneck in the real-time implementation of block matching algorithms is their huge computational load. One of the ways of reducing the complexity is to use a simpler matching criterion than the mean-squared error criteria. In this paper, we propose a matching criterion which is computationally more efficient than the existing criterion. The proposed criterion is similar to the pel difference classification criterion but applied to a reduced number of bits as the reduced bit mean absolute difference criterion. From the simulation results on the displaced frame difference and the displacement vector held, it is expected that motion compensation with the proposed matching criterion will have similar rate-distortion performance as with other existing criteria