An image registration method based on similarity of edge information

Considering that spatial structural information of images, taken from the same scene even with diverse illustration or in different wavebands, has much in common with each other, we propose an image registration method utilizing similarity of edge information as a new similarity measure. Similarity of edge information can be achieved from gradient information of detected edge pixels in two images, so it reflects well the relevance of spatial structures between two corresponding scenes. Experimental results indicate that, no matter when lighting changes or for images sensed in different wavebands, the proposed similarity measure exhibits a perfect single peak response with respect to transformation parameters and the proposed method achieves more accurate alignment than the one based on mutual information.