Application Research Of Computational Ghost Imaging And Deep Learning In Computational Optical Image Authentication And Optical Cryptography

Nowadays,it is an era of information explosion and with the rapid development of network technology.It becomes much easier for hackers to steal personal information which results in widespread leakage of confidential private information,so that the information security(IS)technology which ensures the safety and integrity of the target information attracts wide attention and has been continuously concerned.While the optical information security technology,particularly,has unique advantages in the field of optical image encryption and authentication because of the multidimensional large capacity,high processing speed,and high parallelism of optical transformation.Since 1995,the double random phase encoding(DRPE)optical encryption scheme was first proposed,during the past 30 years,various optical security schemes have been promoted,the optical security technology has been fully studied and has become an inseparable part of modern information security technology.Among them,optical image encryption based on computational ghost imaging(CGI)technology proposed in 2010 broadens the field of optical information security field.Compared with traditional optical encryption methods,the CGI-based optical cryptosystem has simpler and cheaper experimental device,demands less ciphertext space,so that it has aroused wide research interest.On the other hand,computer and network technology has its explosive development during the past two decades,along with the popular of the artificial intelligence(AI)and deep learning(DL).The DL neural networks have solved several problems,which are difficult via traditional method in different fields due to their superior performance.The neural networks are also applied in optical security filed,including image steganography and encryption system,compression and depression during image encryption,robustness enhancement and cryptanalysis of optical cryptosystem et al,which open up a new page of the optical information security technology.This paper takes the computational ghost imaging optical encryption technology as the main line.The first chapter systematically introduces the basic theory of optical information security,including the involved cryptography knowledge and common optical cryptosystems.The second chapter gives a complete description of the origin of computational ghost imaging,including traditional entangled ghost imaging and pseudothermal light source ghost imaging,their experimental devices and corresponding imaging principles are also involved.Then particularly,the main line.computational ghost imaging is introduced,and also the optical encryption scheme based on it.The following work in this paper will be carried out on the CGI-based cryptosystem.The third chapter,based on the singular value decomposition ghost imaging(SVDGI)and the(t,n)threshold secret sharing algorithm,proposes a multi-level image authentication scheme:the key can be divided into n sub-keys and distributed to n users to share the image authentication system.In the system,more than or equal to t authorized users together can realize the high-level authentication of the system and obtain the fully reconstructed certification image;while less than t users,such as only a single one,can only pass the low-level authentication of the system with a meaningless recovered image and obtain lower authority.The SVDGI technology used in the system aims to gain fast and high-quality restoration of the certified image,and the correlation coefficient with the standard original image of the certification center can reach a perfect 1.0(in the simulation experiment).Both simulation experiments and optical experiments have successfully verified the feasibility of the authentication system.Besides,the sensitivity and security of the keys of the authentication system are also analyzed and tested.This work is expected to provide new ideas and inspirations for optical image authentication.Chapter 4,for the CGI encryption scheme,we propose a known-plaintext cryptanalysis system based on the deep learning method of generative adversarial network with physical prior conditions.The neural network proposed is composed of a discriminator and a generator.The training of the cryptanalysis network is based on the game confrontation between the discriminator and the generator.After valid training,while ciphertext image inputs into the cryptanalysis network,a high-quality predicted plaintext image can be obtained in real time.Compared with the previous optical cryptanalysis schemes,we use larger size(128 × 128)face images with more details as the attack targets,which is more difficult and complicated.At the same time,a horizontal comparison of cryptanalysis systems based on classic neural network frameworks,including convolutional neural networks(CNN),recurrent neural networks(RNN)and U-Net structures has put forward,shows the superiority of our scheme in decrypting and restoring the complicated targets.Both the simulation and optical experiments verify the feasibility of the proposed method.Finally,the key points and innovative points of the full text are summarized,as well as the insufficiency of the paper.