ABSTRACT
Image registration (IR) involves determining a mapping between the points of a given input image and a reference image. The reference image is constructed from either a single image or by integrating features from multiple images, where a feature is characterized based on user interest to be either an anatomical structure, functionally active region, or something that relates an anatomical structure to a functional region. In this chapter, our focus is on the development of point feature-based methods for IR. In particular, our focus is restricted to methods for intrinsic non-rigid local registration of two-dimensional magnetic resonance Imaging (MRI) brain images of a single subject (intrasubject) taken over time. Our IR methods exploit combinatorial and information-theoretic techniques to partition the input and the reference images into informative segments and establish a segment map, a mapping between these informative segments. This segmentation and the segment map can help explain our IR. We also provide a statistical guarantee on the quality of IR in terms of a p-value, a probability of obtaining an IR better than our method by random chance.
