How to match: to learn or not to learn? (part 2)

The general pipeline of finding image correspondences.
Nearest neighbor strategy. Features from img1 (blue circles) are matched to features from img2 (red squares). You can see, that it is asymmetric and allowing “many-to-one” matches
Mutual nearest neighbor strategy. Features from img1 (blue circles) are matched to features from img2 (red squares). Only cross-consistent matches (green) are retained.
Second nearest ratio strategy. Features from img1 (blue circles) are matched to features from img2 (red squares). For each point in img1 we calculate two nearest neighbors and check their distance ratio . If both are too similar (>0.8, bottom at Figure), then the match is discarded. Only confident matches are kept. Right graph is from SIFT paper, justification of such strategy.
FGINN illustration from the MODS paper https://arxiv.org/abs/1503.02619
Learning to Find Good Correspondences https://arxiv.org/abs/1711.05971

All the matching strategies except the learned one, with toy examples are implemented here https://github.com/ducha-aiki/matching-strategies-comparison

Matching strategies comparison. The best ones are FGINN union and learned matching.

--

--

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store