Lu Guan

An image stabilization algorithm on corner detection and feature block matching

Moving objects may cause local motion interference and have negative effects for the performance of digital image stabilization. The paper presents a novel method based on corner detection and feature block matching with Hausdorff distance. SUSAN operator is used for corner extraction in large scale feature blocks of uniform distribution. The similarity of the two corner sets of feature blocks between two frames is calculated by introducing weighted corner response function into partial mean Hausdorff distance, and the local motion vectors can be obtained. The final effective motion vectors are determined by checking consistency. The experimental results show that the proposed algorithm can stabilize image sequences with heavy local motion interference, and the average PSNR increases 21.03 dB after stabilization.