Convolutional Hashing for Automated Scene Matching

TL;DR

This paper introduces a novel loss function and training scheme for neural network-based binary hashing, significantly improving automated scene matching accuracy and reducing false positives compared to traditional descriptors.

Contribution

It presents the first neural network that surpasses Haar wavelets and color layout descriptors in scene matching, with a new loss function that relates manifold distances to Hamming space.

Findings

Achieved a 100-fold reduction in false positive rate.

Outperformed state-of-the-art Haar wavelets and color layout descriptors.

Significantly increased true positive rate.

Abstract

We present a powerful new loss function and training scheme for learning binary hash functions. In particular, we demonstrate our method by creating for the first time a neural network that outperforms state-of-the-art Haar wavelets and color layout descriptors at the task of automated scene matching. By accurately relating distance on the manifold of network outputs to distance in Hamming space, we achieve a 100-fold reduction in nontrivial false positive rate and significantly higher true positive rate. We expect our insights to provide large wins for hashing models applied to other information retrieval hashing tasks as well.