Point-matching method for remote sensing images with background variation

Finding correct feature correspondence proves to be difficult in the process of image registration, especially for remote sensing images with background variation (e.g., images taken before and after an earthquake or flood) due to significant intensity differences in the same area. A robust and accurate point-matching method, called triangle transformation matching (TTM), is presented to increase the correct matching ratio and remove outliers. First, scale-invariant feature transform (SIFT) is used to extract the point features, and two preliminary point-matching sets can be obtained. Then, the spatial structure information around one point is compared to its corresponding point in the preliminary matching sets to verify whether they are inliers or not. This structure information is based on triangle area representation and it is affine invariant. A spatial consistency measure is used to remove outliers whose coordinates are very similar. Experiments compared with RANSAC, GTM, Bi-SOGC, and HTSC demonstrate the effectiveness of TTM under the conditions of background variation for remote sensing images.