Pair-Relationship Modeling for Latent Fingerprint Recognition

TL;DR

This paper introduces a novel deep learning approach that models the pair-relationship directly between latent and reference fingerprints, improving recognition accuracy by capturing similarity more effectively.

Contribution

It proposes a hybrid deep network that directly models pair-relationships for latent fingerprint recognition, addressing issues of size variability and corruption.

Findings

Outperforms state-of-the-art methods on two databases

Effectively handles variable sizes and corrupted areas of latent fingerprints

Demonstrates significant accuracy improvements in fingerprint recognition

Abstract

Latent fingerprints are important for identifying criminal suspects. However, recognizing a latent fingerprint in a collection of reference fingerprints remains a challenge. Most, if not all, of existing methods would extract representation features of each fingerprint independently and then compare the similarity of these representation features for recognition in a different process. Without the supervision of similarity for the feature extraction process, the extracted representation features are hard to optimally reflect the similarity of the two compared fingerprints which is the base for matching decision making. In this paper, we propose a new scheme that can model the pair-relationship of two fingerprints directly as the similarity feature for recognition. The pair-relationship is modeled by a hybrid deep network which can handle the difficulties of random sizes and corrupted areas of latent fingerprints. Experimental results on two databases show that the proposed method outperforms the state of the art.