基于匹配关键点裁剪（MKPC）的高效准确共视区域定位：一种提升图像匹配性能的两阶段流水线
计算机视觉与模式识别
2023-03-27 v1 机器学习
摘要
图像匹配是计算机视觉中一个经典且基础的任务。在本文中，基于共视区域之外的区域携带信息很少的假设，我们提出了一种匹配关键点裁剪（matching key-points crop, MKPC）算法。MKPC 以高效和准确的方式定位、提议并裁剪关键区域，即共视区域。此外，基于 MKPC，我们提出了一种通用的两阶段图像匹配流水线，其兼容任意图像匹配模型或组合。我们将 SuperPoint + SuperGlue 接入该两阶段流水线进行实验，结果表明我们的方法提升了室外姿态估计的性能。此外，在公平对比条件下，我们的方法在 Image Matching Challenge 2022 Benchmark 上优于 SOTA，该基准是目前最具挑战性的室外图像匹配基准。
引用
@article{arxiv.2303.13794,
title = {Efficient and Accurate Co-Visible Region Localization with Matching Key-Points Crop (MKPC): A Two-Stage Pipeline for Enhancing Image Matching Performance},
author = {Hongjian Song and Yuki Kashiwaba and Shuai Wu and Canming Wang},
journal= {arXiv preprint arXiv:2303.13794},
year = {2023}
}
备注
9 pages with 6 figures. Many experiments have not yet been conducted, the theoretical sections are rather concise, and the references are not adequately comprehensive. This version of the paper is being released to make this work public, and code will also be published soon. We will continue to conduct additional experiments and periodically update the paper