A new computationally adaptive formulation of block-matching motion estimation

This paper presents a new computationally adaptive formulation of the full-search block-matching (FSBM) motion estimation. A key feature of our approach is to progressively adjust the number of computations required per block to the picture variation by applying a data-driven thresholding. Unlike the related methods, the proposed algorithm utilizes a simple threshold updating mechanism which neither requires a large hardware overhead to eliminate redundant search candidates, nor affects the performance of the motion estimation. Compared to the FSBM, it executes four times fewer operations whilst preserving the same quality of results. The algorithm implicitly features array implementation and can be easily realized in VLSI. A supportive hardware design is outlined.