Robust Feature Correspondences from a Large Set of Unsorted Wide Baseline Images

摘要

Given a set of unordered images taken in a wide area, an effective\udsolution is proposed for establishing robust feature correspondences\udamong them. Two major improvements are made\udin our work as follows: firstly, a robust technique is proposed\udfor the self-organization of a large number of images without\udspatial orderings; secondly, a novel wide-baseline matching\udapproach is developed to obtain good correspondences over\udimages taken from substantially different viewpoints. The\udoutput consists of many sets of reliable pair-wise feature correspondences\udwhich are essential in various computer vision\udapplications. Realistic experiments were carried out to evaluate\udthe performances of the proposed method by using a large\udamount of images captured from our university’s campus.