The goal of image registration is the determination of a common coordinate system in which images can be compared or fused on a pixel-by-pixel basis. Chamfer matching is a classical image registration method for segmented features that was suggested by Barrow et al. in 1977 [2]. It has gained wider recognition through the work of Borgefors, who applied the algorithm in a hierarchical way for locating outlines of lakes in satellite images and investigated some of the algorithm's basic properties [7]. The method basically matches a drawing onto an image. The name chamfer matching is derived from the application of the chamfer distance transform in the algorithm [6]. However, at present, the name chamfer matching is used more generally, irrespective of the used type of distance transform. The word "chamfer" stands for a tool to create a groove in, for example, a wooden table surface. If one considers a drawing as a wire, and a distance transform image as a groove, chamfer matching may be described as the process of letting the wire fall into the groove (Fig. 1).

The chamfer matching algorithm consists of the following independent components: image segmentation algorithms for both images to be registered, a distance transform, a cost function, and an optimization algorithm. We will describe the cost function (and the associated distance transform) first, since it illustrates some important features of the chamfer matching algorithm.

0 0

Post a comment