Fingerprint Pattern Recognition And Classification Based On Neural Networks

As a kind of biological characteristic unique to human beings,fingerprint has its own characteristics of fixation,uniqueness.It can be used as an information transmission tool to identify people's identity.The identification of fingerprint has been successfully applied in many fields of human society.For example,the policeman use fingerprint comparison to confirm the identity information of the suspect;the fingerprint is used as the signing agreement or the signing certificate of both parties;the smartphone is unlocked by setting a fingerprint and applied to the power-on password or payment password.doctors may discover some disease by doing researches with fingerprints.Due to the characteristics of the fingerprint itself and its wide range of use,the research and development of fingerprint identification is very important for human society.Since the deep learning and neural networks have been widely studied and put into practical use by scholars and researchers,the research work on image recognition has made a major breakthrough,and the identification of fingerprint images is no exception.In this paper,by using two kinds of neural networks,convolutional neural networks and deep belief networks,as the basic model,and they are optimized and adjusted in various aspects,and finally obtain the network model which is suitable for fingerprint identification and classification.The main work of this paper is as follows:(1)Through the research and study of the development history and model feature of Boltzmann Machines,Restricted Boltzmann Machines and Deep Belief Networks,and build a deep belief network model to train handwritten numbers image data and fingerprint image data respectively.The training of this network also needs to be compared,which includes the adjustment of important parameters and network performance.Finally,the setting of parameter configuration which is more suitable for this deep belief network is determined,and the optimal performance for the recognition work of both handwritten digital image data and the fingerprint image data is obtained.On this basis,combine the convolution structure with the deep belief networks to get a new model.The new model is used to identify and classify the fingerprint pattern.(2)Through the study and comparison of several classical CNN models,which are LeNet-5,AlexNet,CaffeNet and GoogLeNet,a new CNN is built,and the hierarchical structure and parameters of the network are gradually modified in order to obtain a suitable network model for fingerprint image data which is FCTP-Net.The model includes 4 convolutional layers,3 pooling layers and 3 fully connected layers.And study the specific meanings of the parameters of each layer in the network,complete a series of experiments relying on theoretical basis which includes base learning rate,learning rate policy,maximum iteration,momentum,the initialization method,etc.After that,compare the results in many aspects and obtain one network model with the parameter configuration which is more suitable among all.Then the best performance for identification and classification the fingerprint image data is obtained.In addition to the newly proposed FCTP-Net network,comparative experiments are conducted on several classical network models described above,and compare the structure of several network models and the recognition performance of fingerprint image data.After getting the trained model,use the unused fingerprint image data to complete the identification and classification with new model to check the performance of it.