Hybrid Image Registration based on Configural Matching of Scale-Invariant Salient Region Features Xiaolei Huang (1), Yiyong Sun (2), Dimitris Metaxas (1), Frank Sauer (2), Chenyang Xu (2) (1) Division of Computer and Information Sciences, Rutgers University, Piscataway, NJ, USA (2) Imaging and Visualization Department, Siemens Corporate Research, Princeton, NJ, USA - Email Addresses: Xiaolei Huang: xiaolei@cs.rutgers.edu Yiyong Sun: yiyong.sun@scr.siemens.com Dimitris Metaxas: dnm@cs.rutgers.edu Frank Sauer: frank.sauer@scr.siemens.com Chenyang Xu: chenyang.xu@scr.siemens.com - Abstract: We present a novel approach for aligning images under arbitrary poses, based on finding correspondences between image region features. As a hybrid of feature-based and intensity-based methods, our approach uses a small number of automatically extracted scale-invariant salient region features, whose interior intensities can be matched using robust similarity measures. To estimate the parameters of an aligning transformation, previous techniques have focused on finding correspondences between individual features. In this paper, we emphasize the importance of geometric configural constraints in preserving global consistency of individual matches, thus enabling to eliminate false feature matches. In the proposed approach, on top of the candidate individual feature matches, we further use a Configural Matching step to detect a joint (simultaneous) correspondence between multiple pairs of salient region features. This joint correspondence is chosen as the one having maximum likelihood in terms of global image ``alignedness'', and is derived in an Expectation-Maximization framework. We can then use the joint correspondence to recover the optimal transformation parameters. We applied the proposed approach to registering aerial images and medical images of both single and multi modalities, and found it exhibiting robustness to noise, intensity change/inhomogeneity, and occlusion/disocclusion. It also handles partial matching problems naturally. - Current Link to the paper (PDF): http://paul.rutgers.edu/~xiaolei/regionFeature_registration.pdf