Keypoint matching is a core step in many computer vision tasks. Traditional descriptors such as SIFT and ORB are computed from local image patches, but they can be unreliable in scenes with repetitive patterns or weak texture. The new paper introduces CLAD, a method that instead uses global line correspondences to match keypoints.
By leveraging global line information, CLAD aims to provide more robust matching in challenging environments where local patch descriptors lose discriminative power. The approach is positioned as an alternative to patch-based methods, potentially improving performance in scenes that typically confuse SIFT and ORB.
The paper is available on arXiv under identifier 2610.21409.