A novel multistage image registration technique with graph-based region descriptors
Successful image alignment is an essential function for many image processing methods. The geometric and photometric variations between images adversely affect the ability for an algorithm to estimate the transformation parameters that relate the two images. Local deformations, lightning conditions, object obstructions, and perspective differences all contribute to the challenges faced by traditional registration techniques. In this work, a novel multistage registration approach is proposed that is resilient to view point differences, image content variations, and lighting conditions. The proposed method is demonstrated to be effective for registration scenarios involving images of a scene or object before and after a disaster. Robust registration is realized through the utilization of a novel region descriptor which couples the spatial and textural characteristics of invariant feature points. Clusters of invariant feature points are shown to provide more discriminative features than the traditional point descriptors. ^ The multistage method is a hierarchy of registration approaches that takes advantage of feature, intensity and Fourier-based techniques. The three phases include a limited search window method that employs the proposed graph-based region descriptor, a comprehensive approach which fuses intensity and feature-based analysis, and an exhaustive search method which also utilizes the region descriptor. Each successive stage of the registration technique is evaluated through an effective similarity metric which determines subsequent action. The registration of aerial and street view images from pre and post disaster provide strong evidence that the proposed method estimates more accurate global transformation parameters than traditional intensity and feature-based methods. Experimental results involving the mutual information metric confirm the robustness and accuracy of the proposed multistage image registration methodology. Moreover, experimental results show that the proposed graph-based region descriptor offers higher matching accuracy than SIFT, SURF and BRISK descriptors for the test set of images from before and after a disaster.^
Eliza Y. Du, Purdue University, Jianghai Hu, Purdue University.
Engineering, Electronics and Electrical