A fast digital image stabilization based on large-angle rotational motion estimation

In this paper, a new fast digital image stabilization (DIS) algorithm based on large-angle rotational motion estimation is proposed. An image equivalent motion model is derived so as to decouple rotational motion and translational motion estimation for simplification. Based on this model, a novel large-angle rotational motion estimation algorithm based on bit-plane block matching, which does not require to estimate the rotation center, is presented. Then a modified Adaptive Block-selected Matching Algorithm (ABSMA) is used to estimate the translational motion after rotational motion compensation, and followed by the translational motion compensation. Experimental results show that the proposed rational motion estimation algorithm is significantly effective to the arbitrary rotation reaching 10 degrees, while the whole algorithm shows comparable performance with DIS method based on feature matching and the overall algorithm speed can reach 29 fps for the 640 × 480 resolution video sequences.