Feature-assisted search technique for motion estimation

Motion estimation (ME) is one of the most time-consuming parts in video encoding system, and significantly affects the quality of the reconstructed sequences. In this paper, we propose a new fast algorithm, or the Feature Assisted Search Technique for Motion Estimation (FASTME), which is a gradient descent search combined with a new search strategy, namely, one-dimensional feature matching based on selective integral projections (1DFM). 1DFM assists a local search around an initial search center with a low computational complexity. After performing 1DFM, a small diamond search pattern and more compact horizontal and vertical search patterns, are adaptively used according to the result of 1DFM. The proposed algorithm outperforms existing fast algorithms in terms of speed up performance. In addition, its PSNR performance is better or similar to those of existing fast algorithms.