PhD, Machine Learning Engineer at Gladia
June 8th, 2023
The corresponding point coordinates determined by classic image matching approaches define local zero-order approximations of the global mapping between two images. But the patches around keypoints typically contain more information, which may be exploited to obtain a first-order approximation of the mapping, incorporating local affine maps between corresponding keypoints. Several methods have been proposed in the literature to compute this first-order approximation. In this paper we present several modifications of the RANSAC (RANdom SAmple Consensus) algorithm, that uses affine approximations and a-contrario procedures to improve the homography estimation between a pair of images. The a-contrario methodology provides a definition of the soundness of an estimation and allows for adaptive thresholds for inlier/outlier discrimination. These approaches outperform the state-of-the-art for different choices of image descriptors and image datasets, and permit to increase the probability of success in identifying image pairs in challenging matching databases.