Optical Image Encryption Based On Phase Retrieval Algorithm And Sparse Representation

Information technology brings convenience to people,but also brings some information security issues such as information theft and tampering.Optical information encryption technology has received a lot of attention from scholars with its advantages of high speed,multi-dimensionality and strong security.The background and meaning of the research,the present state of domestic and international research on optical image encryption,cryptanalysis of optical security systems and the application of deep learning in optical encryption are introduced in this thesis.Some basic theories and algorithms of optical information security are then presented summarily.In this thesis,an optical multi-image cryptosystem based on interference and sparsity constraints is presented,and security analysis for several optical cryptosystems by using deep learning methods is also proposed.More specifically,the main research of the thesis is reflected in the following areas.(1)An optical image authentication and encryption system by using the optical interference and sparsity constraints is proposed.In the proposed system,the plaintext images are encrypted into two pure phase masks by using optical interference and amplitude-phase retrieval algorithm,respectively.The pure phase masks are divided to two parts by using multiplexing technique,which are respectively as ciphertext and decryption key.In the decryption process,since the decryption keys contain the ciphertext data from binary images by using double random phase encryption,which need to be authenticated before applied to decryption.And the different plaintexts are reconstructed by using different keys.The numerical simulation shows that the proposed approach has high capacity for multiplexing and security.(2)The security of the non-linear cryptosystem with high security is analyzed by building a Dense Net model.The Dense Net model is designed using a densely connected convolutional neural network and combined with U-net,which is trained by "ciphertext-plaintext" pairs generated by the asymmetric optical encryption system.With pre-trained Dense Net,secret images can be successfully derived from ciphertexts without using decryption keys.The robustness of the Dense Net model is verified by ciphertext shearing and adding noise.Results of simulation experiments show that the nonlinear optical encryption system based on the phase recovery algorithm is vulnerable to the Dense Net model.(3)A deep learning framework based on a U-type convolutional neural network(U-net)is proposed to analysis the security of optical interference-based encryption system with sparsity constraints.In the optical cryptosystem,two phase-only masks can be obtained by using phase retrieval algorithm with the constraints of complementary binary amplitude masks.By integrating the interference technique and phase encoding,the encryption scheme achieves a high security level and is immune to conventional attack methods.Here,we investigate and construct a neural network architecture for evaluating the security of the cryptosystem based on the idea of U-net.In this method,the approximate equivalent model of the compressive interference-based optical cryptosystem is achieved by using a U-net model,which has been pre-trained by a series of “ciphertext-plaintext” pairs.The simulation results show that the cryptosystem can be broken by using the proposed deep learning network.The plaintexts can be retrieved from their corresponding ciphertexts without the use of security keys,such as the diffraction distances,the binary amplitude masks,etc.