| Network communication has involved many fields of society.While greatly promoting social development,network communication is facing a serious threat due to the openness of the network environment and the vulnerability of communication data.Information hiding technology can provide security for network communication by protecting communication content and hiding communication behavior.In this paper,aiming at the security problems of the existing information hiding technology,such as abnormal media content,statistical traces and difficult to resist the detection based on high-dimensional statistical characteristics,the generative adversarial network,covert channel and image steganography are studied and improved,so that the undetectability and robustness of information hiding technology are improved.Based on the theory of information hiding,a dynamic key information hiding system for instant communication is designed and implemented.The main work of this paper includes the following aspects :（1）Aiming at the problem that the time-hidden channel is easy to be detected and the robustness of communication process is poor,an orthogonal wassertein generative adversarial network that can recover the input noise is proposed,and the generative time-hidden channel is constructed through this network.Based on wassertein generative adversarial network,the discriminant ability of the discriminator is decomposed into noise recovery ability and discriminant ability according to the model idea of orthogonal generative adversarial network,and the dynamic batch size update strategy is introduced to improve the convergence rate of the model.Based on the time covert channel of generative adversarial network,the covert mapping model of time behavior and covert information of network data stream is established by network model adversarial training.The covert mapping model adaptively hides and recovers secret information through neural network,which significantly improves the information hiding performance of time covert channel.（2）Aiming at the limitation of prior key agreement in existing image steganography algorithms,a general model of dynamic key image steganography is proposed.The model combines time covert channel with image steganography.The key of image steganography is hidden and transmitted through time covert channel,which realizes the dynamic change of key and improves the concealment of image steganography.The model can be applied to most keybased image steganography methods.This paper takes PNG image steganography as an example to introduce the hiding,transmission and recovery process of the model in detail.（3）Based on the general model of dynamic key image steganography,a dynamic key information hiding prototype system is designed and implemented.The system is developed based on Java and python,and a communication client is built on Windows platform.Web Socket and http protocol are used to realize time covert channel communication and secret image transmission.Finally,this paper tests the function and performance of the system.In the process of functional testing,the information hiding function and non-information function of the system can run normally,and the information hiding function runs accurately,which breaks the limitations of traditional image steganography.In the system performance test,the orthogonal wassertein generative adversarial network has strong noise recovery ability and fast model convergence rate,which can meet the needs of information hiding.The temporal covert channel based on the generated adversarial network has good performance in robustness and undetectability;PNG image steganography performs well under SSIM and PSNP evaluation metrics.Although the steganography capacity is limited by the covert channel capacity,it can still reach 255 bytes. |
