Is Hamming distance the only way for matching binary image feature descriptors?

TL;DR

This paper investigates whether alternative metrics to Hamming distance can improve binary image feature matching accuracy, finding some metrics perform better but without statistically significant differences.

Contribution

The study evaluates non-Hamming metrics like Jaccard-Needham and Dice for binary descriptor matching, providing a statistical comparison of their effectiveness.

Findings

Jaccard-Needham and Dice metrics sometimes outperform Hamming distance

Performance differences are not statistically significant

Hamming distance remains a robust choice for binary descriptor matching

Abstract

Brute force matching of binary image feature descriptors is conventionally performed using the Hamming distance. This paper assesses the use of alternative metrics in order to see whether they can produce feature correspondences that yield more accurate homography matrices. Two statistical tests, namely ANOVA (Analysis of Variance) and McNemar's test were employed for evaluation. Results show that Jackard-Needham and Dice metrics can display better performance for some descriptors. Yet, these performance differences were not found to be statistically significant.