Research On Three-dimensional Fingerprint Recognition Based On OCT

Optical coherence tomography(OCT)is a new three-dimensional tomography technology developed gradually in the 1990 s.It has a bright application prospects in biological tissue detection and imaging.Based on the principle of low coherence interference,OCT can obtain the tomographic capability in depth direction.It can reconstruct two-dimensional or three-dimensional images of the internal structure of biological tissues or materials.Traditional two-dimensional fingerprint technology has been successfully applied to fingerprint recognition based on skin surface texture information,such as sweat,texture and the like.However,it is susceptible to changes in the skin's epidermis,including abrasion and cutting,corrosive burns,and plastic surgery.In this paper,optical coherence tomography(OCT)is applied to obtain high-resolution images of epidermal and dermal fingerprints to construct three-dimensional fingerprints.The results show that the OCT-based 3D fingerprint extraction technology can extract the surface features and internal features of fingerprints.The technique can identify the texture features of the dermis layer fingerprint,and the dermis layer texture feature can be matched with the epidermis layer,and the dermis layer texture information collected by OCT is used to match the epidermis information by using the neural network classification algorithm,thereby achieving the effect of identifying the identity and greatly improving the effect.The efficiency and limitations of identification.In this thesis,a fingerprint recognition technology based on three-dimensional reconstruction is proposed.In fact,the three-dimensional fingerprint recognition technology based on three-dimensional reconstruction is not to acquire the three-dimensional structure of finger skin texture,but to expand the three-dimensional fingerprint model to two-dimensional plane,so that the expanded results can be matched based on the existing two-dimensional fingerprint matching algorithm,and can be compatible with the existing two-dimensional fingerprint database.Finally,the process of matching and matching eigenvalues is realized,and the fingerprint feature recognition ofreal three-dimensional structure is realized.High resolution images of epidermal and dermal 3D fingerprints were obtained using optical coherence tomography.A matching technique of dermis and epidermis based on the acquired three-dimensional fingerprint,that is,a classification neural network matching algorithm,extracts a block region range and enhances an image by calculation.The matching strategy is to first reconstruct the three-dimensional image collected by the OCT,extract the deep leather fingerprint of the image,splicing the leather fingerprint and the skin fingerprint,and input the image—the convolution layer,the nonlinear layer,the pooling layer(down sampling),Fully connected layer-output classification-gives the probability of matching.The test fingerprint library of this experiment comes from volunteers,and the fingerprints used are matched and numbered,and the identity information is not reflected.The experiment randomly selected 18 people as the source of the sample,and each person collected 10 finger fingerprint images.The fingerprinting process is performed during the collection process,such as finger watering,coloring,and the like.Usually,a certain proportion is used to randomly extract fingerprint images for training and testing.The size of the original image is 482×432.The classified CNN model used in this paper is 7hidden layers.The activation function of each layer adopts the sigmoid function.In the image input layer,the size of the image is uniformly adjusted to 227×227,which facilitates the subsequent feature extraction.The main research content as follows:1.This topic is based on the construction of the OCT system.Therefore,it is necessary to first build a reliable OCT system.The experiment built a fiber optic OCT system based on the Michelson interferometer structure,with a center band of1310 nm near-infrared light.The system's highest axial scanning(A-scan)speed is76 KHZ,the actual axial resolution is 12?m,and the largest imaging is 5.8mm,enabling real-time high-resolution imaging of biological tissue.The system runs at a sampling rate of 20 frames per second.Each frame contains 1000 A-scans and can be3 D reconstructed to obtain a spatial 3D image of the sample.2.For human epidermal fingerprints are susceptible to changes in skin epidermis,including wear and cutting,corrosive burns and plastic surgery,as well as changes in physical conditions,such as finger dirt,wet fingers.In the case where the conventional fingerprint scanner cannot extract the surface fingerprint.This paper proposes to use OCT to collect 3D fingerprints to obtain dermis layer texture information,to achieve the purpose of identifying identity,and to combine the deep learning classification neural network CNN to match the epidermis and dermis.The basic steps are as follows: In step 1,the image is simply resized so that each image input to the input layer is the same size.In step 2,the input layer is followed by the convolution layer.The function of the convolutional layer is to filter the local part of the image by several convolution kernels with different weights.The number of convolution kernels is the number of different features that can be obtained at the same position.Therefore,a convolution kernel can get a Feature Map.In step 3,the convolutional layer is usually followed by a down sampling layer(not absolute),which further reduces the size of the feature map by a maximum pooling or average pooling operation and maintains the rotation and shifting of the feature to some extent.Not denatured.In the fourth step,after several layers of convolutional layer and the down sampling layer are processed,the picture is classified and identified by a full connection operation.