Research On Fingerprint Liveness Detection In Meteorological Information Management System

At present,China is in the critical period of vigorously promoting the development of meteorological informationization,and the meteorological information management system with the core of meteorological data sharing applications often faces security problems such as illegal intrusion and access by users.To improve system security,varieties of information protection technologies are proposed.A password is the most commonly used authentication method for obtaining access to a meteorological information management system,but it is easily leaked,forgotten,or deciphered.Fingerprints are used as an identity authentication certificate because of their uniqueness and stability.However,in actual applications,illegal users can invade the meteorological information management system through artificial fingerprints,which brings security risks to system access.Therefore,this thesis will study the fingerprint liveness detection algorithm to resist the fake fingerprint spoofing attack,and apply the fingerprint liveness detection technology to the fingerprint authentication system of the meteorological information management system to further ensure the authenticity of the user identity of the meteorological data.The main research contents are as follows:(1)A fingerprint liveness detection algorithm based on spatial texture feature extraction has been proposed.The artificial fingerprint can reproduce the coarse-grained texture information in the true fingerprint image,so the fingerprint recognition system in the weather information management system is vulnerable to the attack of the fake fingerprint.However,it is difficult for fake fingerprints to duplicate high-quality subtle textures,making illegal certification possible.Gradient can reflect the intensity of the pixel variation in the image,which is good for characterizing the small-scale texture features in the fingerprint.Thus,this paper first proposes a feature extraction method of multi-directional differential co-occurrence matrix,which first pre-processes and quantizes the image,and then calculates the horizontal and vertical gradients.Secondly,this thesis designs a truncation operation method.By analyzing the relationship between the gradient and the truncation factor T,the problem of the outliers due to the interference of illumination intensity and noise are eliminated,and the accuracy of the algorithm is improved.Finally,the co-occurrence matrix features of the difference matrix in four different directions are calculated,and the normalized features are fed into support vector machine(SVM)to train and test.Additionally,two fingerprint identification systems that support fingerprint liveness detection are presented in chapter 7.(2)A fingerprint liveness detection algorithm based on feature extraction of transform domain system is proposed.The spectral distribution in the transform domain helps to reveal the image texture characteristics that are not observed in the spatial domain,and extracting the significant difference between the true and false fingerprints can improve the performance of fingerprint liveness detection in the meteorological information management system.Wavelet transform can highlight the features of the image in scale and direction,and is very suitable for complex and variable fingerprints.In this thesis,we first use the first-order wavelet transform to transform the texture analysis of the image from the spatial domain to the frequency domain.By analyzing the transform domain coefficients,two feature extraction algorithms of the transform domain coefficients are proposed.One is a feature extraction algorithm based on local binary mode.Firstly,the window is selected and slides in a fixed step size to compare the relationship between the center of the window and the adjacent pixels,and encode,then,the processed image is divided into blocks and characterized.The other is the feature extraction method of multi-order difference co-occurrence matrix,which is to calculate the distribution relationship of two adjacent,three or four gradients and use it as the feature vector.Finally,the normalized features are fed into SVM.Additionally,this thesis gives the relational expression between the matrix order and feature dimension.Experiments show that the proposed algorithm has better detection performance in the transform domain.(3)A fingerprint liveness detection method based on convolutional neural network model is proposed.The use of meteorological high-performance computers can shorten the running time of the deep learning algorithm,train the true and fake fingerprint models offline and transplant them into the fingerprint liveness detection of the meteorological information management system,which can further improve the accuracy of the system to detect true and false fingerprints.This thesis first proposes a scale equalization convolutional neural network(CNN)model to solve the constraints of CNN on the input image scale.CNN is prone to slow convergence or local optimality when back-propagating update parameters.The thesis introduces a learning rate adaptive adjustment module to optimize it.Moreover,this thesis solves the problem of ineffective area interference in fingerprint images by designing a region of interest extraction algorithm.To make the model classifier have better generalization,this thesis designs a fingerprint liveness detection algorithm based on local gradient pattern texture enhancement,and has achieved good results on the Liv Det 2011 dataset.To solve the problem that the training parameters face gradient expansion or gradient disappearance,and reduce the training time of the model in the back propagation process and avoid the parameters falling into the local optimum,this section also proposes a deep residual network model based on adaptive learning.(4)A fingerprint liveness detection algorithm based on deep multi-modal feature fusion is proposed.Learning the significant difference between true and fake fingerprints is the key to improving the performance of fingerprint liveness detection in meteorological information management systems.The traditional detection methods are based on the characteristics of manual design.In order to make full use of deep learning,the characteristics of the original image structure and texture can be automatically learned from the tagged data.In this thesis,the network before the convolutional neural network output layer is used as feature extraction.A fingerprint liveness detection algorithm for deep feature self-learning is proposed.Then,to test the performance of feature fusion under different modes,this thesis designs a multi-modal depth feature fusion method.Before the feature extraction,two operations of transfer learning and model fine-tuning are respectively performed to solve the problem of insufficient training set,and the fine-tuned training parameters are used as the initial values of the feature extractor.Finally,the fused features are fed into the classifier to train and test.In addition,this thesis also tested the fusion features on two classic classifiers,extreme learning machine and softmax,and obtained good detection performance.