Graph-Theoretic Image Alignment Using Topological Features

In this paper, we introduce a feature-based image alignment method using topological singularities. The main idea behind our proposed framework is to encode a medical image into a set of Morse critical points. Then an entropic dissimilarity measure between the Morse features of the target and the reference images is maximized to bring the data into alignment. We also show that maximizing this divergence measure leads to minimizing the total length of the joint minimal spanning tree between the features of the misaligned medical images. Illustrative experimental results clearly show the much improved performance and the registration accuracy of the proposed technique.