Compensating for sub-pixel shift in motion estimation

In block-based sub-pixel motion estimation, most methods are based on a comparison of the target block to a number of sub-pixel shifted versions of a reference block. This means that the two blocks, reference and target, are not treated equally, as one is takes as-is, while the other suffers sub-pixel shifting, which invariably results in loss of information and degradation. In this paper we propose a method for sub-pixel motion estimation that treat reference and target block identically, and calculates sub-pixel motion via a third block, constructed on the spot, from which the reference and target blocks may be approximated by appropriate sub-pixel shifts. Tests on a number of test sequences show that the proposed method performs much better than standard sub-pixel motion estimation.