Motion vector optimization of control grid interpolation and overlapped block motion compensation using iterative dynamic programming

Interdependence between motion vectors (MVs), introduced by control grid interpolation (CGI) and overlapped block motion compensation (OBMC) algorithms, is the key to improving temporal prediction performance of conventional block-matching motion compensation schemes. Unfortunately, this dependency makes the problem of finding optimal MVs intractable. While standard schemes that successively optimize each MV are susceptible to severe local minimum problems, we propose a dynamic programming (DP) paradigm, where each horizontal or vertical slice of MVs is jointly determined during an iterative optimization process. To retain reasonably low complexity, our algorithm effectively identifies an initial search region and then chooses a proper search scheme for each MV. In addition, a computationally-efficient multiscale search strategy is employed. The performance of the proposed method is compared with that of the standard optimization techniques, and our experimental results show that the proposed scheme always gives a better rate-distortion performance. Especially for CGI, the PSNR improvements and the percentage of bit-rate savings provided by our algorithm, in some cases, are in excess of 1.0 dB and 20%, respectively.