【作者】 胡海洋；

【导师】 谭光华； 孙斗；

【作者基本信息】 湖南大学 ， 计算机技术（专业学位）， 2018， 硕士

【摘要】 指纹识别技术目前已广泛用于门禁打卡、设备解锁、移动支付以及第二代身份证验证等场景之中。利用指纹等生物特征进行身份验证的优点在于个体生物信息的独特性,另外一方面它们使用简单并且难以复制。指纹活体检测在过去许多年里一直是一个十分活跃的研究课题,目前已经证明可以通过标准的光学和电容传感器进行假冒欺骗。基于指纹的认证系统需要一个有效的方法来对真假指纹进行正确的区分。针对此课题,本文从传统纹理及目前流行的深度学习两方面对指纹活体检测进行研究。本文主要内容如下:(1)原始获取的指纹图像往往具有噪声且指纹图像中包含大量空白的前景图像,本文提出在进行指纹活体检测时首先利用设计的指纹图像裁剪算子进行图像裁剪;接着进行去噪等预处理;为了利用噪声分量等尖锐信息对指纹活体检测的贡献,在得到导向滤波去噪后图像的同时也综合使用到原始裁剪后的图像;最后利用CoALBP特征提取算子对两类图像进行特征提取。实验结果表明,提出的检测方法能够有效的检测出真假指纹,提升了指纹识别系统的安全性。(2)认识到目前基于卷积神经网络等深度学习的方法在图像识别领域的巨大优势;本文使用卷积神经网络在指纹活体检测上进行迁移应用,对指纹图形的真伪进行二分类识别。本文同时结合传统计算机视觉图像预处理的方法,首先对指纹图像进行裁剪和去噪处理;在此基础上,对处理后的图像分块并使用提出的网络模型进行训练;最后,利用训练好的模型对单张测试图像的分块进行类别判断,使用投票策略最终确定图像类别。(3)设计和实现一个指纹活体检测仿真系统,在提升指纹识别系统检测性能和安全性方面展现出了较好的应用场景。更多还原

【Abstract】 Fingerprint liveness detection has been widely used in access control,cell phone un-locking,payment and the second generation ID card.Using biometric features such as fingerprint to authenticate takes advantage of the uniqueness of individual biological infor-mation.Iris and fingerprints are unique to everyone.Besides,they are simple to operate but difficult to duplicate.The fingerprint liveness detection has been an active research topic in the past few years,and it has been proved that the standard optical and capacitive sensors can be used for counterfeiting and deception.Fingerprint based authentication system needs to create a way to distinguish between genuine and fake fingerprints.To solve this problem,this paper studies fingerprint detection from two aspects:traditional texture and current popular deep learning.The main contents of this paper are as follows:(1)The original fingerprint image often has a lot of noise and the fingerprint image contains a large number of blank foreground images.For the problems stated above,this paper advocates to using the proposed fingerprint image cutting operator to cut the finger-print image first,then carrying out denoising and other preprocessing.In addition,in order to make use of the contribution of the sharp information such as the noise component to the detection of the fingerprint,not only guided filtering but also the cutting image is used to do denoising,then the CoALBP feature extraction operator is used to extract the features of the two types of images.The experimental results show that the proposed detection method can effectively detect the genuine and fake fingerprints,and enhance the security of the fingerprint identification system.(2)Realizing the great advantage of the convolution neural network in the field of image recognition,this paper uses the convolution neural network to carry out the migra-tion application on the fingerprint liveness detection,and classifies the fingerprint into two classifications.At the same time,combined with the traditional computer vision image pre-processing method,the fingerprint image is first tailored and denoised.On this basis,the processed image is partitioned and trained using the proposed network model.Finally,the model,which has been trained last step,is used to judge the blocks of the single test image and the voting strategy is used to ultimately determine the whole image’s class.(3)A fingerprint detection simulation system has been designed and implemented.It has shown a better application scene in improving the detection performance and security.更多还原