Abstract

Pore detection for fingerprint recognition has gained much research attention in recent years, in view of the existence of large number of pores in a small fingerprint segment and availability of high-resolution acquisition devices. Current research efforts have focused on developing two-part hybrid schemes, wherein the first part is comprised of a CNN architecture to produce a pore intensity map and the second part consists of a scheme that determines the pore centroids exploiting the knowledge base on the pores characteristics using this pore intensity map. However, CNN architectures used in the first part of the existing pore detection schemes are unable to extract pore features that adequately represent the fingerprints at a reasonable computational cost and in the second part the methods are not able to exploit the knowledge base on fingerprint pores efficiently. In this paper, a new two-part fingerprint pore detection scheme is proposed, wherein the first part focuses on developing a CNN architecture capable of extracting highly representational pore features and the second part on accurately determining the pore centroids by taking into consideration the inadequacies in fingerprint acquisition and distinguishing the spatial characteristics of true and false pores. Extensive experiments are performed to demonstrate the distinct characteristics to show the superiority of the proposed scheme in performance and complexity over the existing state-of-the-art pore detection schemes.