Learning a Fixed-Length Fingerprint Representation

Abstract

We present DeepPrint, a deep network, which learns to extract fixed-length fingerprint representations of only 200 bytes. DeepPrint incorporates fingerprint domain knowledge, including alignment and minutiae detection, into the deep network architecture to maximize the discriminative power of its representation. The compact, DeepPrint representation has several advantages over the prevailing variable length minutiae representation which (i) requires computationally expensive graph matching techniques, (ii) is difficult to secure using strong encryption schemes (e.g. homomorphic encryption), and (iii) has low discriminative power in poor quality fingerprints where minutiae extraction is unreliable. We benchmark DeepPrint against two top performing COTS SDKs (Verifinger and Innovatrics) from the NIST and FVC evaluations. Coupled with a re-ranking scheme, the DeepPrint rank-1 search accuracy on the NIST SD4 dataset against a gallery of 1.1 million fingerprints is comparable to the top COTS matcher, but it is significantly faster (DeepPrint: 98.80% in 0.3 seconds vs. COTS A: 98.85% in 27 seconds). To the best of our knowledge, the DeepPrint representation is the most compact and discriminative fixed-length fingerprint representation reported in the academic literature.

Privacy-Preserving Fingerprint Authentication Using D-H Key Exchange and Secret Sharing

“…Recently, some novel fingerprint representations demonstrated superior authentication accuracy, which lay a great foundation for the future development of fingerprint template protection methods, for example, the Minutia Cylinder-Code (MCC) [32], a representation based on 3D data structures (called cylinders) and the DeepPrint [33], a fixedlength fingerprint representation of only 200 bytes. Built upon this novel fingerprint representation, we will investigate template protection for fingerprint biometric systems.…”

C2CL: Contact to Contactless Fingerprint Matching

“…Scores are computed using both of these representations and then fused together using a sum score fusion for a final similarity score. 1) Texture Representation: To extract our textural representation, we fine-tune the DeepPrint network proposed by Engelsma et al in [28] on a training partition of the publicly available datasets which we aggregated (Table III). We note that we do not include any data from our newly collected ZJU dataset for fine-tuning as we want this dataset to remain completely unseen for a more rigorous evaluation.…”

“…Our approach is more comprehensive than published methods which focus on only a subset of these modules needed for state-of-the-art performance. 2) A fully automated, preprocessing pipeline to map contactless fingerprints into the domain of contact-based fingerprints and a contactless-contact adaptation of Deep-Print [28] for representation extraction. Our preprocessing and representation extraction is generalizable across multiple datasets and contactless capture devices.…”

Training Neural Networks on Top of Support Vector Machine Models for Classifying Fingerprint Images

“…Militello et al [42] showed the performance of pre-trained CNNs, including AlexNet [31], GoogLeNet [55], and ResNet [26] for the fingerprint image classification. DeepPrint network [21] learns alignment and minutiae from fingerprint images, making fixed-length fingerprint representations of only 200 bytes. Kai and Anil [5] proposed a CNN to learn an orientation field dictionary for fingerprint alignment.…”