Alignment of challenging image pairs: Refinement and region growing starting from a single keypoint correspondence Gehua Yang Charles V. Stewart Michal Sofka Chia-Ling Tsai Our goal is a registration algorithm capable of aligning image pairs having some combination of low overlap, large illumination differences (e.g. day and night), substantial scene changes and different modalities. Our approach starts by extracting and matching keypoints. Rankedordered matches are tested individually in succession. Each is used to generate a transformation estimate in a small image region surrounding the keypoints. The growth process works by iterating three steps: 1) refining the estimate by symmetrically matching features on the two images, 2) expanding the region according to the uncertainty in the mapping, 3) selecting an appropriate transformation model. Image features are corner points and face points located by analyzing the intensity structure of image neighborhoods. After convergence, if a correctness test verifies the transformation it is accepted and the algorithm ends; otherwise the process starts over with the next keypoint match. Experimental results on a suite of challenging image pairs shows that the algorithm substantially out-performs recent algorithms based on keypoint matching. Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY cs-05-13
Alignment of challenging image pairs: Refinement and region growing starting from a single keypoint correspondence
Gehua Yang
Charles V. Stewart
Michal Sofka
Chia-Ling Tsai
Our goal is a registration algorithm capable of aligning image pairs having some combination of low overlap, large illumination differences (e.g. day and night), substantial scene changes and different modalities. Our approach starts by extracting and matching keypoints. Rankedordered matches are tested individually in succession. Each is used to generate a transformation estimate in a small image region surrounding the keypoints. The growth process works by iterating three steps: 1) refining the estimate by symmetrically matching features on the two images, 2) expanding the region according to the uncertainty in the mapping, 3) selecting an appropriate transformation model. Image features are corner points and face points located by analyzing the intensity structure of image neighborhoods. After convergence, if a correctness test verifies the transformation it is accepted and the algorithm ends; otherwise the process starts over with the next keypoint match. Experimental results on a suite of challenging image pairs shows that the algorithm substantially out-performs recent algorithms based on keypoint matching.
Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY
cs-05-13