Hierarchical Pore-Based High-Resolution Fingerprint Indexing

Pore-based fingerprint recognition has become more attractive in recent years because of its uniqueness and difficulty in forgeability. However, most of the existing methods use pore features for fingerprint verification rather than fingerprint indexing, and most of them use only local features of pores to evaluate the similarities between fingerprints. In this article, we present a hierarchical pore-based high-resolution fingerprint indexing system using two kinds of pore features (including pore-based global and local features). Our work consists of two major parts. First, we design a united model named deep pore-based global and local features model (DPGL) to extract pore-based global and local features of fingerprints simultaneously. The most important is the DPGL framework makes full use of pore information. It applies deconvolution module to learn a spatial attention mask that explicitly exploits self-supervision on pore locations. Second, we design a novel hierarchical pore-based fingerprint indexing approach. In the proposed approach, the reference fingerprints are primely selected based on pore-based global features. The fingerprints retained are selected based on pore-based local features subsequently. Finally, the candidate fingerprints are ranked according to pore information matching. Experiments on two datasets show that DPGL not only can accomplish the extraction of pore-based global and local features successfully, but also can achieve self-supervision on pore information. Compared with other state-of-the-art pore-based fingerprint indexing methods, the proposed approach reduces the pre-selection error rate by about 50% at penetration rate 1%, and performs at least 65% reduction in retrieval time.