Hierarchical Clustering in Face Similarity Score Space

TL;DR

This paper explores hierarchical clustering of face images based on similarity scores, revealing meaningful groupings by subject, gender, ethnicity, and illumination, aiding in understanding face recognition data structures.

Contribution

It introduces hierarchical clustering methods applied to face similarity score spaces, uncovering multi-level taxonomies including race, gender, and illumination conditions.

Findings

Clusters align with race, gender, and illumination.

Hierarchical methods effectively organize face similarity data.

Evidence supports meaningful subgroupings in face recognition datasets.

Abstract

Similarity scores in face recognition represent the proximity between pairs of images as computed by a matching algorithm. Given a large set of images and the proximities between all pairs, a similarity score space is defined. Cluster analysis was applied to the similarity score space to develop various taxonomies. Given the number of subjects in the dataset, we used hierarchical methods to aggregate images of the same subject. We also explored the hierarchy above and below the subject level, including clusters that reflect gender and ethnicity. Evidence supports the existence of clustering by race, gender, subject, and illumination condition.