Study On Iris Image Encryption And Liveness Detection Algorithm Based On Deep Learning

In modern cryptography,the secret key,a guarantee of the algorithm security,is the core information to encryption and decryption.To ensure image security,many image encryption algorithms were proposed,which are often related to the key security.However,the security of key transmission and management is still a problem in the research of cryptographic algorithms.As a new encryption technology,the biometric encryption technology generates the key based on individual biological characteristics,a basis keyless encryption of information.Iris has excellent encryption characteristics,i.e.,uniqueness,stability,non-invasion,rich information and strong anti-attack ability,which is suitable for information encryption.The iris image encryption algorithm becomes a new research branch of the image encryption technology and plays an important role in image encryption.In this thesis,the deep learning algorithm is used to extract iris image features.The key is extracted from the extracted biometric features.Two kinds of encryption algorithms based on the the deep learning and an iris liveness detection algorithm are proposed.The main contributions are as follows:(1)To facilitate the key management,an iris image encryption algorithm based on the deep learning is proposed.This algorithm trains deep learning model with iris data sets.After image preprocessing,image blocks regarded as stacked auto-encoder input are extracted to learn parameters.By adding the interval constraints,the parameters learned are used to initialize the parameters of convolutional neural network.The 256-dimension vectors are extracted by the trained deep learning model,and the key is generated by Reed-Solomon error-correcting algorithm.The key is used to encrypt and decrypt the image by advanced encryption standard.The Reed-Solomon code is used to reduce the influence of the environment on the iris.The encryption accuracy and security analyses show that the new algorithm can decrypt the original image effectively and has high security.(2)To facilitate the key management,a keyless image encryption algorithm based on the deep learning is proposed.This algorithm extracts the same feature vectors from different individuals with the deep learning model.the key is generated by Reed-Solomon code.The original image is encrypted by advanced encryption standard with the generated key.In this algorithm,the encrypted key is generated by the sender Alice's iris,and the decrypted key is generated by the recipient Bob's iris.To realize the successful decryption,we view Alice's data set and Bob's data set as one class data to train the deep learning model.After that,the Reed-Solomon code is used to reduce the influence of the environment on iris.Through the analysis of the decryption accuracy and security of the proposed algorithm,the new algorithm greatly reduces the probability that the key is compromised,and effectively realizes the image encryption and decryption.(3)To resist the fake-iris attack,a multi-module iris detection algorithm based on the feature fusion is proposed.The development of industrial technology promotes fake iris to attack the iris recognition system.To prevent the loss caused by fake iris attack,a multi module feature fusion iris activity detection algorithm based on the original algorithm is improved.First,the iris features are extracted by the deep learning;second,the texture features of iris images are extracted with the local binary pattern and Gabor filtering algorithm,with importance of texture features and a large amount of information considered;third,fuse the texture features with iris images,and set a threshold to eliminate the image blocks with poor quality;finally,the iris liveness is detected is by the convolutional neural network and the classifier.Experiments were carried out on several iris data sets,and the comparisons are also made.Experimental results and algorithm analyses show the feasibility of the new algorithm.