Fusion of Minutia Cylinder Codes and Minutia Patch Embeddings for Latent Fingerprint Recognition

TL;DR

A fusion based local matching approach towards latent fingerprint recognition that improves rank-1 identification accuracy by considerably for real-world datasets when compared to either the single usage of these features or existing state-of-the-art methods in the literature.

Abstract

Latent fingerprints are one of the most widely used forensic evidence by law enforcement agencies. However, latent recognition performance is far from the exemplary performance of sensor fingerprint recognition due to deformations and artifacts within these images. In this study, we propose a fusion based local matching approach towards latent fingerprint recognition. Recent latent recognition studies typically relied on local descriptor generation methods, in which either handcrafted minutiae features or deep neural network features are extracted around a minutia of interest, in the latent recognition process. Proposed approach would integrate these handcrafted features with a recently proposed deep neural network embedding features in a multi-stage fusion approach to significantly improve latent recognition results. Effectiveness of the proposed approach has been shown on several public and private data sets. As demonstrated in our experimental results, proposed method improves rank-1 identification accuracy by considerably for real-world datasets when compared to either the single usage of these features or existing state-of-the-art methods in the literature.

Figures (7)

• Figure 1: This figure illustrates an overview of the proposed feature level fusion approach towards latent fingerprint recognition task. MCC cylinder vectors and MinNet embedding vectors are generated around each minutiae (shown in red dots) for both query (latent) and target (sensor) images. Next, 2 separate similarity score matrices are generated corresponding to MCC cylinders and MinNet embeddings, and top matching minutiae pairs are selected from these similarity score matrices. Finally, we concatenate these matching minutiae pairs for 2 different descriptors and generate a match score using Local Similarity Sort with Relaxation (LSS-R) algorithm.
• Figure 2: Sample images from private EGM Test Dataset (first-row) and JGK Test Dataset (second row). Pairs of rolled (left) and latent (right) images from EGM and dataset; (a)-(b) and (c)-(d).
• Figure 3: FVC Latent Test Dataset Examples a-c) Slap Fingerprint b-d) Fake Latent Fingerprint
• Figure 4: This figure presents a sample latent-rolled pair image from EGM Test Dataset. A visual illustration of the best matching 8 minutiae pair correspondences for (left) MinNet MinNet2022, (middle) $MCC_{FM}$ and (right) $P_{feature}$. While MinNet MinNet2022 and $MCC_{FM}$ is able to match in rank-1 and rank-2, respectively, there exists a false matched minutiae correspondence shown in red line.
• Figure 5: Cumulative Match Characteristic (CMC) curves on EGM Test Dataset for top-3 performing methods listed in Table \ref{['tab:EGM_results']}.