Implementation of a Fingerprint Recognition System Using LBP Descriptor

Abstract

Abstract Conventional fingerprint recognition systems provide authentication by a direct matching of minutiae points and orientation field. Although several resemblance algorithms have been proposed, reliable automatic fingerprint verification remains a challenge due to the difficulty in alignment for direct matching and the construction of adequate functions for resemblance measurements. In this paper, we propose a solution to the aforementioned problems using a local binary pattern (LBP) descriptor applied to minutiae and orientation fields. The experimental results on the public fingerprint database, Fingerprint Verification Competition (FVC), show high recognition rates. The proposed system was implemented on the platform known as FPGA Virtex-II Xilinix™ (Virtex2p-xc2vp7-FF672) and optimized with respect to hardware resources occupation, based on a co-design methodology. All the proposed algorithms are involved in the design of a mixed software/hardware dedicated system. A classifier based on pulse mode neural networks using floating-point format is proposed.