Main problems and proposed solutions for improving template matching

Abstract

In this work we discuss the problems of template matching and we propose some solutions. Those problems are: 1) Template and image of search differ by a scale, 2) Template or image of search is object of rotation, 3) Template or image of search is object of an affinity. The well known method is NCC (Normalized Cross Correlation); this method can not handle scale, rotation, affinity or occlusion. Also the NCC is not preferred for binary image. So we propose here to use index similarity for example Jaccard index.