An Image Stabilization Algorithm Based on Feature Matching Applicable to Large Rotation

A new stabilization algorithm is presented, which can solve the problem of the translational and large rotational motion between image sequences simultaneously. Based on corner point detection, matching strategy featured point-to-multipoint is presented. And a new parameter extraction algorithm is deduced. The algorithm has made a breakthrough to the limit of most of the previous algorithms´ capability of only resolving translation and small rotation, and it has also overcome a drawback of the shortcoming of log-polar transform based algorithm that the center of rotation must be precisely determined. The experimental results show when the rotation angle is 18° the algorithm can achieve image stabilization. After image stabilization the average PSNR is enhanced about 26 dB.