A flexible algorithm for image matching

We propose a method for measuring the similarity between grey level images. The method is able to match images successfully even in the presence of small geometric deformations, illumination changes, and severe occlusions. It fits naturally an implementation based on a comparison of data structures which requires no numerical computations. The range of its applications is vast, and in particular it is a useful tool for object detection and iconic search. We present very good results on real images with and without occlusions, and a qualitative comparative study with a well-known correlation method.