Robust and effective techniques for multi-modal image registration

Image registration is the process of estimating the optimal transformation that aligns different imaging data into spatial correspondences. Multi-modal image registration is to register images which are captured by different types of imaging devices. This thesis aims to develop robust and effective techniques for multi-modal image registration. The challenge lies in the fact that the visual appearance may differ a lot between corresponding parts of multi-modal images. We have been exploring ways by investigating local image features. Two main contributions have been made in this thesis. First, we have improved existing mono-modal and multi-modal image registration techniques by better utilizing gradient information. For a feature-based image registration technique, its effectiveness to a large extent relies on the discrimination power of local descriptors. In the existing techniques, gradient information is utilized in a number of ways for building local descriptors. We have analyzed the limitations of these techniques, and have proposed a technique for better utilizing gradient information. As a result, the discrimination power of local descriptors has been enhanced, leading to a better registration performance. Second, we have developed a new multi-modal image registration technique, which has the following innovations: 1. We have proposed a technique to detect the intrinsic structural similarity in multi-modal microscopic images. This is achieved by exploiting the characteristics in intensity relationships between the Red-Green-Blue color channels. 2. To increase robustness to content differences, contour-based corners are used, instead of intensity-based keypoints in a state-of-the-art multi-modal image registration technique. 3. We have proposed a new local descriptor to better represent corners. 4. We have proposed a new way of scale estimation by making use of geometric relationships between corner triplets in two images. The proposed multi-modal image registration technique achieves greater robustness in terms of both content differences and scale differences as compared to the state-of-the-art multi-modal image registration technique.