Nonlinear Similarity Based Image Matching

Image matching is an inarguably important operation for many practical sophisticated systems in machine vision and medical diagnosis. Many gray-level image matching applications use the sum-of-squared-difference (SSD) or sum-of-absolute-differences (SAD), which are very sensitive to noise. Almost all images have some kind of noise, which causes the matching tasks significantly difficulty. In this paper we explore a new, less noise sensitive image-matching technique. It uses non linear similarity measure min or median on interest points to find a match. The algorithm has been tested using a range of images with different gaussian noise. The result shows a significant improvement over traditional Euclidean distance measure technique for image matching.