Optimum selection algorithm of motion estimation blocks for fast and robust digital image stabilization
H. Okuda, Manabu Hashimoto, Kazuhiko Sumi, Kazunori Sasaki · 2003
We propose new optimum selection algorithm of motion estimation blocks to improve accuracy and computational cost. In our algorithm, motion vectors are segmented robustly to some regions which support each motion model. The experimental results show that it achieved the high accurate of motion estimation, and the computing time was shortening up to approximately 1/20 compared with the conventional hierarchical full-search method. It is confirmed that our algorithm has enough robustness under ill-conditioned scenes.