Cryptanalysis Of Optical Cryptosystem Based On Genetic Algorithm And Deep Learning

With the rapid development of information technology,the security of information and data is becoming more and more serious.Optical cryptography based on optical theory and method,as a new type of information protection technology,has the advantages of parallel data processing and high security performance,attracting more and more scholars to participate in it,and various optical cryptography technologies have been put forward one after another.For any kind of optical cryptographic technology,it must be strictly analyzed for security before it can be put into use.Therefore,security analysis of some typical optical cryptosystems can not only effectively avoid the loss caused by security vulnerabilities,but also promote the development of optical cryptography technology.This paper first summarizes the recent development of optical cryptography and optical cryptanalysis,and introduces the concept of cryptography and the basic theory of optical information processing technology.For optical cryptosystems based on random phase coding,two new optical cryptographic analysis methods based on genetic algorithm and deep learning are proposed in this paper.The core contents and innovations of the thesis can be summarized in the following two aspects:(1)A choose plaintext attack method based on genetic algorithm is proposed.The proposed method aims at double random phase encoding system.By using genetic algorithm to optimize the modulation of the optical field phase in frequency domain,the "equivalent key" of the cryptosystem is obtained,and the confusion and diffusion of the random phase key in the system are successfully removed.With this "equivalent key",an attacker can directly decipher plaintext information from arbitrarily intercepted ciphertext.Compared with the existing selective plaintext attack method,this method does not need to record any complex amplitude distribution and has higher robustness to noise.(2)An optical cryptanalysis method based on deep learning is proposed.The proposed method designs a deep neural network for optical cryptanalysis.By using many pairs of "ciphertext-plaintext" images to train the network,the trained network can be regarded as the "equivalent key" of the cryptosystem and can be used to decipher the plaintext information.This method is not only applicable to the traditional double random phase coding system,but also to the triple random phase coding system with higher security.The practical optical experiment results further verify the feasibility and practicability of the method.As an application exploration of deep learning in optical cryptanalysis,the proposed method provides a new idea for the security analysis of optical cryptosystems.