Robust Image Registration Based on Features Extracted from Image of Edges

This paper describes a registration method of images taken from two different parts of the electromagnetic spectrum: the infrared and the visible range. Image registration includes development of the transformation in order to achieve an optimal overlapping of image pairs. The transformation is affine and consists of scaling along the x- and y-axis, translation, rotation and horizontal skew, and it is found based on features composed of two characteristic points (intersection of linear segments and/or corners) connected with one linear segment. These features are detected and extracted from the both images of edges (IR and RGB). As a measure of the best overlap, i.e., the optimal transformation, a function is used which takes into account the number of corresponding points in both images and the sum of their distances. The method is experimentally verified on numerous pairs of infrared and visible images of building facades. The examples of image registration and its evaluation are given. Key words: image registration, feature detection, feature extraction