Optical Image Encryption Based On Ghost Imaging

With the development of information and communication technology,huge volume of data is transmitted over internet.Resultantly,the issue of information security becomes more and more important.Cryptography plays a very important role in data security,which basically hides the sensitive information so that it cannot be read by anyone except the person whom we want to send it.Optical encryption has been developed rapidly because of its advantages of high encryption efficiency,multiple coding dimensions and high-speed parallel processing capability.Recently,the optical encryption system based on ghost imaging has attracted wide attention because of its low requirements on imaging hardware and smaller ciphertext storage space.However,the application of ghost imaging encryption is still restrained from time-consuming measurements and the large number of measurements.In this thesis,compressive sensing and deep learning methods are used to reduce the number of measurements and improve the quality of reconstructed images.In addition,two new encryption methods were proposed as follows:1.A ghost imaging method is proposed based on deep learning,by which one can obtain high quality approximations of object images directly from the one-dimensional detected bucket signals.The implementation of efficient image reconstruction for ghost imaging is performed by an end-to-end generative adversarial network,which is pretrained with simulated data.An ‘‘U-net’’ based architecture is used as the generator of the proposed deep neural network and a convolutional classifier is employ as the discriminator to learn the features of the one-dimensional bucket signals generated by conventional ghost imaging method.The effectiveness of the proposed scheme is experimentally verified by using two types of targets.To further investigate the performance of our method,we compared it with traditional and compressive sensing ghost imaging approaches,U-net-based ghost imaging respectively.2.A multiple-image encoding,compression and authentication mechanism is proposed to provide multi-level encryption and multi-level compression.Before proceeding to compression and encryption using ghost imaging,the target images are digitally compressed and encoded into a pseudo-random image based on Walsh-Hadamard transform and Arnold scrambling transformation.The resultant image is further encoded into one sequence base on a ghost imaging method.In the decryption procedure,after the operations of compressive sensing,inverse Arnold scrambling transformation and inverse WalshHadamard transform,the original images can be successfully reconstructed from cyphertext sequence.In our proposed method,an authentication mechanism has been introduced to facilitate the pairing of cyphertext and keys and reduce the decryption time when the incorrect decryption keys are provided,which is important but can be easily missed in general ghost-imaging-based security systems.The execution of decryption must be terminated if the authentication fails.3.A color computational ghost imaging encryption system based on deep learning is designed,in which steganography is introduced to improve the security level.In the encryption procedure,the detected intensities of the secret image are hidden into those of a non-secret image by means of non-obvious data combination,and the combined data is used as the ciphertext of steganography.In the decryption procedure,the embedded secret data needs to be extracted from the ciphertext and correlated with the color speckle and the resultant data is then input into the multi-discriminator generative adversarial network.If the ciphertext is directly decoded without the operation of extraction of the secret data in advance,only the non-secret information can be restored rather than the secret information,which implies that the scheme can defend against the attacks from deep learning.Besides,we further introduce an index key in the process of data integration to enhance the security of the steganography scheme.