Digital Image Stabilization Based on Circular Block Matching

Abstract

In this paper, a novel digital image stabilization algorithm based on circular block matching is proposed to estimate the global scaling, rotational and translational motion parameters between each two consecutive frames in jitter videos. In the algorithm, the rotation-invariant-features based circular block matching is proposed for the first time to estimate the local true motion vectors for the centers of all considered circular blocks. With these local motion vectors and their corresponding block positions, a linear system is constructed and solved by repeated least-squares to generate the global motion parameters (GMP) which are then refined by a gradient descent based iteration scheme. The experimental results demonstrate that the proposed DIS technique can generate more precise GMP and deal with larger motions compared with the ever presented optical flow based digital image stabilization technique.