Authors: Badreddine Griouz; Rafik Djemili; Hocine Bourouba; Hakim Doghmane
Addresses: PI:MIS Lab, Université 8 Mai 1945 Guelma, BP. 401, Algeria 24000, Guelma, Algeria ' LRES Lab, University of 20 Août 1955, Skikda, Algeria ' PI:MIS Lab, Université 8 Mai 1945 Guelma, BP. 401, 24000, Algeria ' PI:MIS Lab, Université 8 Mai 1945 Guelma, BP. 401, 24000, Algeria
Abstract: Finger vein patterns have been proved as one of the most promising biometric modality for its convenience and security. Most of the current available finger vein recognition methods utilise features from a segmented blood vessel network. This manner of processing however may not provide optimal recognition accuracies as reported in many studies. Therefore, this paper proposes in the feature extraction stage, the use of the spatial pyramid decomposition (SPD) method aiming at partitioning the finger vein images into increasingly fine subregions from which local texture descriptors are obtained. The descriptors adopted in this paper are local binary pattern (LPB), binarised statistical image feature (BSIF) and local phase quantisation (LPQ). The performance of the proposed approach evaluated on two publicly databases PolyU and SDUMLA achieves a recognition accuracy higher than that of some existing systems reported in the literature for both the SDUMLA and the PolyU databases.
Keywords: finger vein recognition; FVR; spatial pyramid decomposition; SPD; local binary pattern; LBP; binarised statistical image feature; BSIF; local phase quantisation; LPQ.
International Journal of Biometrics, 2020 Vol.12 No.2, pp.131 - 146
Received: 04 Sep 2018
Accepted: 04 Jun 2019
Published online: 02 Jun 2020 *