Research Of End-to-end Fingerprint Template Protection Algorithm Based On Deep Learning

Biometrics provides a new mechanism for identity management because of its higher security and greater convenience.However,the number of biometric is limited and largely unchangeable.Once the storage database is attacked,it will cause serious security problems such as privacy disclosure.Secondly,as one of the most commonly used biological characteristics,fingerprint database is irreparable once lost due to its uniqueness and lifelong invariance.Therefore,fingerprint template protection technology has received more and more attention and research.At present,the area of fingerprint collection is getting smaller and smaller due to the constant update of convenient equipment,which leads to less effective details in the collection area,which limits the accuracy of the existing template protection methods.From the perspective of image processing,this paper studies the end-to-end deep learn-based fingerprint template protection method,and makes some innovative work in the fingerprint template protection neighborhood.The main contents and contributions include the following parts:1)This paper proposes an end-to-end fingerprint feature extraction network based on deep learning based on neural network.This network model can not only realize fingerprint recognition,but also carry out feature matching of fingerprint image encryption domain.In this paper,the self-built FS200 conventional fingerprint database and the internationally recognized FVC2002 DB1 fingerprint database were tested.The experimental results were as follows: the EER(equal error rate)of the matching results of FS200 and FVC2002 DB1 conventional fingerprint database were 1.01% and 1.68%,respectively.2)This paper improves the fingerprint image template protection algorithm based on pixel mixing-washing,enhances the security of pixel interpolation,and generates hash check code and pixel mixing-washing template to form a dual security mechanism,which effectively improves the anti-similarity attack performance.The algorithm is more robust than the traditional pixel matching algorithm.In this paper,a self-built small area fingerprint library is used to carry out experiments on this algorithm.Results: EER was 2.52%.3)This paper proposes a fingerprint template protection algorithm based on depth fingerprint features,which extracts the original fingerprint template features into 512 d feature vector strings based on deep learning,and then converts the 512 d real feature vector strings into 512 d binary strings using random mapping and nonlinear multidimensional spectral hash transform.In this paper,a self-built small area fingerprint library is used to carry out experiments on this algorithm.Results: EER was 2.13%.Compared with the traditional linear multi-dimensional spectral hash transform,this algorithm has better anti-similarity attack performance and higher accuracy than the 2)algorithm.4)This paper designs and develops a fingerprint file encryption and decryption system based on Fuzzy Vault,and realizes the function of encryption and decryption of files by using fingerprint binding random key string combined with AES encryption mode.In summary,the two template protection algorithms proposed in this paper are superior to the traditional classic small area fingerprint template protection algorithm.At the same time,the results of the end-to-end fingerprint matching algorithm based on deep learning are comparable to the results of the traditional fingerprint matching algorithm,which promotes to a certain extent Application of fingerprint image matching and template protection.