A research on multisensor image registration based on edge matching and point matching

摘要

Multisensor images registration is the necessary premise for multisensor image fusion in military field. The goal of image registration is to establish the geometrical correspondence among images which are acquired by different sensors or by the same sensor but in different modalities. In this paper, a research on multisensor image registration is done and two methods are presented. One method is based on edge matching, the other is based on point matching. Experimental results with various kinds of image have verified the robustness of the two methods. It is showed that point-based method works well for image pairs with large scale and rotation transformation while the gray level characteristics is well preserved, such as same sensor images but different times. Edge-based method works well for image pairs with small scale and rotation transformation while edge characteristics can be well preserved, such as multisensor or multispectral images. Both methods are automatic and computationally quite efficient. Root mean square error at the control points are given and outperformed by manual registration.