Image registration based on geometric pattern matching

摘要

We present an image registration method based on geometric pattern matching. The method is applicable to the situation that there is a similarity transformation between two images. First of all, the edges of the images were detected and the edge points were described with subpixel position and direction of gradient vector of intensities. The geometric de-noising method was proposed to smooth the position and direction of the edge points. And then a coarse registration was performed to compute the similarity transformation parameters with pixel accuracy. In the end, a high accuracy registration based on iterative closest point (ICP) algorithm is used to obtain the final result.