Versatile spectral methods for point set matching

This work is concerned with the problem of point set matching over features extracted from images. A novel approach to the problem is proposed which leverages different techniques from the literature. It combines a number of similarity metrics that quantify measures of correspondence between the two sets of features and introduces a non-iterative algorithm for feature matching based on spectral methods. The flexibility of the technique allows its straightforward application in a number of diverse scenarios, thus overcoming domain-specific limitations of known techniques. The proposed approach is tested in a number of heterogeneous case studies: of synthetic nature; drawn from experimental biological data; and taken from known benchmarks in computer vision.

Research highlights
► The work develops a novel, non-iterative, approach to the point set matching problem over features extracted from images.
► The flexibility of the technique allows its straightforward application in a number of scenarios, overcoming domain-specific limitations.
► The techniques is tested against a number of cases, showing computational efficiency and high quality matching.