MFFFLD: A Multimodal-Feature-Fusion-Based Fingerprint Liveness Detection

Biometrics spoofing attack (BsSA) frequently occurs when an adversary impersonates a lawful user to access to the biometric system by means of some forged or synthetic samples, especially in fingerprint or face authentication. In allusion to the problem above, the mainstream countermeasure, called biometrics liveness detection (BLD), is raised. In this article, we propose a more robust and accurate BLD strategy by taking advantage of weighted multimodal convolutional neural networks (MCNNs) to extract diverse deep features. Before detection, the ROI operation first is performed to remove those invalid backgrounds of fingerprints. Then, a multimodal feature fusion strategy is proposed to make full use of the learning capacity of convolutional neural networks (CNNs) without human interactions. It is well known that characteristics of the different direct splicing together and for the subsequent classification is unreasonable, thus, a weighted summation strategy is explored. More specifically, we assign each type of feature weight contribution rate, sum them, and then learn the optimal combination of different model features. In the final detection phase, to verify our proposed algorithm with higher accuracy, detailed analyses of the fingerprint evaluations on intradatabase, cross-material, and cross-database, respectively, which also include assessment under the fusion of different modal features, and face evaluations on the same-database, are evaluated. Experimental results on several benchmark data sets LivDet 2011, 2013, 2015, and NUAA demonstrate that our approach achieves outstanding results in fingerprint intradatabase and cross-material evaluations as well as face anti-spoofing evaluations comparing with previous methods. Most importantly, our method is more accurate and robust than other existing fingerprint anti-spoofing methods when evaluating cross-database.