Research On Image Steganography Based On Deep Neural Network

The content of digital images is rich and colorful,and is widely used in many fields such as business,politics,and military.The rapid development of cloud computing and the internet has led to the storage and transmission of a large number of images on the network.However,these images usually contain private information,important corporate data,and military confidential information.Therefore,preventing the leakage of image information has become an imminent problem.At present,deep neural networks have achieved good results in many fields,and gradually penetrated into the field of image steganography,and showed its potential application value.This article mainly studies from two aspects: cover modification and coverless information hiding.Aiming at the information hiding of cover modification,image steganography based on deep convolutional neural network is proposed.The hidden network and the extraction network are built with deep convolutional neural networks.The distribution of steganography images is close to the distribution of original cover images,and the secure transmission of secret images is achieved by transmitting steganography images.Aiming at coverless information hiding,the coverless image steganography based on WGAN-GP(Improved Training of Wasserstein GANs)model is proposed to realize the secure transmission of secret images by transmitting unmodified disguised images.(1)Aiming at the problems of low information hiding capacity and weak ability of anti-steganalysis,this paper proposes the image steganography scheme based on deep convolutional neural network.First,the convolutional neural network is used to construct a hidden network and an extraction network.In order to improve the steganography effect,a residual block is added to the hidden network.Then,the sender inputs the secret image and the cover image into the hidden network to generate a steganography image with the secret image hidden.Then finally,the recipent receives the steganography image and inputs it into the extraction network.The extraction network extracts the secret image from the steganography image and displays it.This method not only has a good effect in resisting the detection of steganalysis algorithm,but also improves the capacity of information hiding.(2)In order to effectively avoid the detection of steganalysis algorithms and improve the security performance of information steganography,a research scheme of coverless image steganography based on WGAN-GP model is proposed.In this method,the WGAN-GP model needs to be constructed,and the transmitted disguised image is input to the generation network,and the original secret image is input to the discrimination network.Then,the disguised image and secret image are used to conduct adversarial training on the WGAN-GP model until the image generated by the generative model is the same as the original secret image,the generative model is saved,and the generation model database is constructed.Finally,use the disguised image and the corresponding generation network model to generate the secret image.In this method,the disguised image is transmitted without any modification,which will not cause the suspicion of the attacker during the transmission process,realize the secure transmission of the secret image and increase the information hiding capacity.