Optical Image Encryption Based On Single-pixel Imaging And Deep Learning

There are many kinds of information security problems such as information leakage,theft and tampering in the age of information.Thus,how to keep information safe during the storage and transmission of information has become a very important issue in the research of information processing.Notably,optical image encryption techniques have drawn widespread attention due to their intrinsic properties of parallel processing and multi-dimensional coding.On the one hand,single-pixel imaging(SPI)has natural characteristics of encryption in the process of optical information acquisition.On the other hand,deep learning(DL)has shown apparent advantages in image reconstruction.Hence,optical image encryption based on SPI and DL become a hot spot of research in the field of optical information security.The combination of SPI and DL provides a new approach to further ensure the security of image information and high-quality image reconstruction.In this thesis,we first introduce the background and significance of optical image encryption,and status of domestic and international research on SPI encryption.We then present the basic principles of SPI and DL,as well as some reconstruction algorithms.The main aim of our work is to provide new optical security schemes based on SPI and DL,and the main achievements presented in this thesis are given as follows:(1)We propose a steganographic optical image encryption based on SPI and an untrained neural network.In this encryption scheme,random binary illumination patterns are projected onto a secret image and light intensities reflected from the image are then detected by a bucket detector.To enhance the security of collected secret data,a steganographic approach is introduced in this method,which implements data hiding with a SPI system using encoded illumination patterns.A non-secret image is illuminated with a sequence of encoded patterns that were generated from the scrambled measurements of the secret image,and sequential cyphertext data can be obtained by collecting the diffraction data with the bucket detector.Different from traditional SPI-based encryption schemes,an untrained neural network is adopted as a SPI-encrypted image processor,which allows to reduce time spent on data preparation and reconstruct the secret images with high quality.(2)Providing secure and efficient transmission for multiple optical images has been an important issue in the field of information security.Here we present a hybrid image compression,encryption and reconstruction scheme based on deep learning-assisted SPI and orthogonal coding.In the optical SPIbased encryption,two-dimensional images are encrypted into one-dimensional bucket signals,which will be further compressed by a binarization operation.By overlaying orthogonal coding on the compressed signals,we obtain the ciphertext that allows multiple users to access with the same privileges.The ciphertext can be decoded back to the binarized bucket signals with the help of orthogonal keys.To enhance reconstruction efficiency and quality,a DL framework based on Dense Net is employed to retrieve the original optical images.(3)An optical image hiding scheme based on complex amplitude modulation and SPI is proposed.In SPI systems,a series of random illuminated patterns are used as the key to encrypt the hidden image.The measurements collected by a bucket detector are numerically processed to obtain the binary sequence.The complex amplitude information is then encoded into a modulated pattern of a digital micromirror device(DMD)using the superpixel scheme and the binary sequence.In the decryption,the optical reconstruction image of the DMD pattern is used as the key for information authentication.The illumination patterns are transmitted to the receiver once the reconstructed image is authenticated.The high-fidelity image can be directly reconstructed by inputting illumination patterns and bucket signals decrypted from the DMD pattern into an untrained neural network.