基于纹理特征融合的指纹活性检测方法

Texture feature fusion based fingerprint liveness detection

针对当前指纹识别系统容易遭受伪造指纹欺骗攻击的问题,提出一种基于纹理特征融合的指纹活性检测算法。通过设计边缘纹理增强(ETE)和对称差分统计(SDS)2种脊线纹理特征描述算子来表示真假指纹的显著性纹理,前者用来提取指纹图像脊线的方向纹理信息,后者用来描述邻域内脊线的频率纹理信息。首先,利用感兴趣区域(ROI)提取算法对指纹图像进行预处理,以消除指纹图像中背景空白噪声的干扰;然后,利用ETE和SDS分别提取指纹的脊线纹理特征;接着,统计上述2类特征的直方图,描述真假指纹的纹理特征;最后,将生成的特征输入支持向量机(SVM)中进行训练和测试。在LiveDet 2011指纹数据集的测试中,分别使用Biometrika、Italdata、Sagem 3种传感器,且与Best、韦伯局部描述算子(WLD)、局部相位量化(LPQ)和局部二值模式(LBP)4种指纹检测算法进行了比较,该文算法的检测性能优于其余方法,能够完成当前的活性检测任务。LiveDet 2013数据集使用Biometrika、Italdata和Swipe 3种传感器,通过与WLD、不变梯度直方图(HIG)、统一局部二值模式(ULBP)、深度表征结构优化(DRAO)和Winner 5种指纹活性检测方法对比,该文算法的指纹活性检测准确率有一定的提升。

To solve the problem that the fingerprint recognition system is vulnerable to spoofing attack of forged fingerprints,a fingerprint liveness detection algorithm based on texture feature fusion is proposed.Two ridge texture feature descriptors,edge texture enhancement(ETE)and symmetric difference statistics(SDS)are designed to represent the salient texture of true and fake fingerprints.The former is used to extract the direction texture information of the ridge of fingerprint image,while the latter is used to describe the frequency texture information of the ridge in the neighborhood.Firstly,the region of interest(ROI)extraction algorithm is used to preprocess the fingerprint image to eliminate the interference of background blank noise.Then,the ridge texture features of fingerprints are extracted by ETE and SDS respectively.Secondly,the histograms of ETE and SDS are used to describe the texture features of true and fake fingerprints.Finally,the generated features are input into the support vector machine(SVM)for training and testing.In the testing of the LiveDet 2011 fingerprint dataset,three sensors,Biometrika,Italdata and Sagem are used,and this method is compared with four fingerprint detection algorithms,Best,Weber local descriptor(WLD),local phase quantization(LPQ)and local binary patterns(LBP).The proposed algorithm outperforms other methods in terms of detection performance and is able to complete the current activity detection task.The LiveDet 2013 dataset uses three sensors:Biometrika,Italdata and Swipe.Compared with five fingerprint activity detection methods:WLD,histograms of invariant gradients(HIG),uniform local binary pattern(ULBP),deep representations architecture optimization(DRAO)and Winner,the proposed algorithm has a certain improvement in fingerprint activity detection accuracy.