Modified Ransac for Sift-Based Insar Image Registration

Abstract

Abstract-In this paper, we propose a modified version of the Random Sample Consensus (RANSAC) method for Interferometric Synthetic Aperture Radar (InSAR) image registration based on the ScaleInvariant Feature Transform (SIFT). Because of speckle, the "maximization of inliers" criterion in the original RANSAC cannot obtain the optimal results. Since in InSAR image registration, the registration accuracy is in inverse proportion to number of residues. Therefore, we modify the old criterion with a new one -Minimization of residues -to obtain the optimal results. We tested our method on a variety of real data from different sensors, and the experimental results demonstrated the validity and robustness of the proposed method.

Feature Point Matching with Matching Distribution

“…Although using the local characteristics of feature points can sometimes lead to lots of false match, there are several algorithms to clear up these false matches such as RANSAC [17]. RANSAC was used in [9,18] to improve the performance of SIFT-based feature point matching and the results are impressive.…”

“…Additionally, a subset of matches can be selected according to some conditions, such as angular [7] and pairwise constraints [8], to get better matching results. A random selection can also be applied to improve the results [9]. Unfortunately, this approach cannot provide good matching results since the same pattern geometric structure might be found in any part of the images.…”

Efficient method for outlier removal in SAR image matching based on epipolar geometry

“…Some methods that automatically remove falsely matched points use random sample consensus (RANSAC) [16–18]. The RANSAC automatically determined the coefficients of energy function [18] or a transformation function [12, 19] and has the best performance in removing false matches. Phase, as well as amplitude, information can be used for evaluating the accuracy of the matched points [20].…”