Research On Image Steganography Algorithm Based On Style Transfer And Generative Adversarial Networks

Nowadays,in the era of Internet of Things,the network is full of a lot of sensitive information,and the secure transmission of confidential information is faced with severe challenges.As deep learning and convolutional neural network become a hot topic in computer vision research,steganography researchers combine them with image steganography algorithms and achieve some success.Deep learning and convolutional neural networks have been shown to be able to extract robust features from hidden images.Based on deep learning and convolutional neural network,this paper proposes two image steganography schemes and designs an image steganography system as follows:(1)A carrier-free image steganography algorithm based on style transfer is proposed.The algorithm uses the color style of the secret image to generate the camouflage image and transmits it on the public channel.The algorithm is composed of two parts,namely,the process of generating the camouflage image and reconstructing the secret image.In the course of creating the camouflage image,we adopt the style transfer technique,which combines the style of the secret image and the public image to produce the camouflage image.In the process of secret image reconstruction,a convolutional neural network(CNNSI)is designed to reconstruct the secret image.The training data set of CNNSI network is composed of camouflage images under various attacks.This method does not need to modify the pixel value of the carrier image and can resist common attacks.Experiments show that the proposed algorithm is robust and secure,and it can be used to extract the hidden image even when it is under attack.(2)A steganographic algorithm(CIS-GAN)based on GAN network is proposed to modify the carrier image.In this algorithm,the game idea of generating antagonistic network is introduced,the embedded network with embedded secret image and the extraction network with extracted secret image are regarded as generators,and the steganographic analysis network is introduced as the discriminator.Through continuous training and optimization of steganographic model,the security of steganographic algorithm is improved.In addition,an attack layer is added to the CIS-GAN model,which is mainly used to reproduce multiple attacks suffered by the loaded image during transmission on the channel,so as to improve the resistance of the loaded image to attacks.Finally,a large number of experiments prove that this method can effectively reduce the detection rate of steganographic analysis algorithm,and this method can still effectively extract the secret image after the loaded image is attacked,with good robustness and security.(3)According to the research content of this paper,a digital image steganography system is designed and implemented.The functions of this system include three modules:carrier-free image steganography,embedded image steganography and image processing.The designed system realizes the function of hiding and extracting confidential information and can guarantee the secure communication of confidential information.