Ranking list preservation for feature matching. (March 2021)

• Record Type: Journal Article
• Authors: Jiang, Junjun; Ma, Qing; Jiang, Xingyu; Ma, Jiayi
• Is Part Of: Pattern recognition. Volume 111(2021)
• Publication Date: 2021-03
• Subjects: Feature matching -- Mismatch removal -- Top K rank similarity -- Local neighborhood structure; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2020.107665
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We present a very simple but very effective robust feature matching algorithm. We are the first to utilize the ranking list to represent the local neighborhood structure. The top K ranking distance is used to calculate the local neighborhood similarity. A closed-form solution of the objective function is deduced. It is robust to a large proportion of outliers, and achieves a good precision/recall balance. Abstract: Feature matching plays a very important role in many computer vision and pattern recognition tasks. The spatial neighborhood relationship (representing the topological structures of some key feature points of an image scene) is generally well preserved between two feature points of an image pair. Several mismatch-removing methods that maintain the local neighborhood structures of potential true matches have been proposed. Defining local neighborhood structures is a crucial issue in the feature matching problem. In this paper, we propose a robust and efficient topological structure measurement called top K rank preservation (TopKRP) for mismatch removal from given putative point set. We transform feature points from the feature space to the ranking list space. Thus, the topological structure similarity of two feature points can be simply calculated by comparing their ranking lists, which are measured by the top K ranking similarity based on the spatial Euclidean distance as well as the angle correlation. TopKRP is validated on 10 public image pairs with typical scenes and 2 artificially established datasets, namely, MI52 and RS153 . Experimental results demonstrate that the proposed approach outperforms several state-of-the-art feature matching methods, especially when the number of mismatches is large.