OCT External And Internal Fingerprint Extraction Method Based On Fully Convolutional Neural Network

Fingerprint recognition is one of the most important technologies in biometrics.At present,fingerprint recognition has been widely used in criminal investigation,finance,exit-entry and other fields.However,when a fake fingerprint is used,the conventional automatic fingerprint recognition system is easily spoofed,resulting in security problems.The papillary layer under the finger skin has an internal fingerprint structure that is identical to the external fingerprint.It is the source of the external skin fingerprint and cannot be easily counterfeited by a fake fingerprint.Optical Coherence Tomography(OCT)is a non-invasive high-resolution imaging technology that clearly captures the internal fingerprint information of a finger and enhances the anti-counterfeiting ability of the fingerprint.In this thesis,OCT technology is used to acquire subcutaneous three-dimensional data of fingerprints.An external and internal fingerprint extraction algorithm is proposed based on improved fully convolutional neural network.The work is summarized as follows:(1)Most existing algorithms to extract these contours are not precise.Based on the internal structural characteristics of the subcutaneous fingerprint,we propose an improved fully convolutional neural network,combining convolutional long-term memory network,residual connection structure and batch normalization layer to the original fully convolutional neural network.Through the network,the three-dimensional data of the OCT finger is directly segmented,and the stratum corneum and papillary layer contour are obtained.Experiments show that using the proposed algorithm to extract the contours of the stratum corneum and papillary layer is better than other traditional algorithms and neural network algorithms.(2)We use the grayscale features and relative distance features of the stratum corneum and papillary layer for generating OCT external fingerprints and internal fingerprints with different methods.Experiments are performed to evaluate the image quality of these methods for generating fingerprints.Thereafter,gray correction is performed by Retinex algorithm for the gray unevenness problem of some OCT fingerprint images.The Gabor filtering is used to enhance the ridges of the fingerprints.Finally,we extract the minutiae of the fingerprint and remove the fake minutiae.Experiments show that the method can accurately extract the minutiae of most of the OCT external fingerprints and internal fingerprints,and prepare for the subsequent fingerprint recognition work.(3)OCT fingerprint data has high anti-counterfeiting ability.Observing the three-dimensional visualization and the slice image of normal fingerprint and fake fingerprint,we find that the structure has huge difference and can judge fingerprint authenticity easily.It proves that the OCT three-dimensional finger data has high anti-counterfeiting ability.This thesis implements OCT external fingerprint and internal fingerprint extraction algorithm based on improved fully convolutional neural network.The experiment proves the validity of OCT external and internal fingerprints and the high anti-counterfeiting of OCT finger three-dimensional data.