基于深度卷积神经网络的指纹活体检测算法研究

Research on a Fingerprint Liveness Detection Algorithm Based on Deep Convolution Neural Networks

近年来,随着指纹认证系统的广泛应用,伪造指纹检测受到人们的日益关注。文章结合卷积神经网络在计算机视觉、人脸识别和图像分类领域的应用特征,提出一个指纹活体检测算法F-net。算法采用BN层、inception结构和全局均值池化层对网络进行优化,以减少F-net网络的大量参数和计算复杂度,这也使得算法在采用大学习率进行网络训练的同时获得了较高的泛化能力。文章在LivDet2011和LivDet2013数据集上进行了多种算法测试。实验结果表明,文章提出的F-net具有较高的泛化能力和实时检测性能。

With the wide application of fingerprint authentication system in recent years, forged fingerprint detection has been paid more and more attentions. Based on the application characteristics of convolutional neural network in the fields of computer vision, face recognition and image classification, this paper proposes a fingerprint liveness detection algorithm called F-net. The algorithm uses BN layer, inception structure and global mean pool level to optimize the network, so as to reduce the large number of parameters in F-net network and the computational complexity. This also makes the algorithm get a higher recognition rate when the algorithm uses a large learning rate to train the network. Many algorithms are tested on LivDet 2011 and LivDet 2013 datasets. The experimental results show that F-net has high recognition rate and real-time detection performance.