Research On Coverless Image Steganography Method Based On Deep Learning

Image steganography is one of the methods used for the hidden exchange of information.It is the art and science of invisible communication,which strives to hide the existence of the communicated message.In the face of increasingly powerful steganalysis,image steganography based on modification is difficult to resist the detection of steganalysis.By not modifying the original digital image carrier,coverless image steganography hides the information in the carrier,which effectively improves the security of image steganography.The emergence of deep learning provides a new way for coverless image steganography.coverless image steganography based on deep learning has high research value,and its application prospects are relatively broad.The problem of coverless image steganography based on deep learning is studied.The characteristics of the attention mechanism are analyzed.Attention mechanism and generative adversarial network are combined.A generative image steganography method based on an attention mechanism is proposed.The structure of the generative adversarial network is studied.The structure of the double branch encoder is designed.With unsupervised learning,the dual-branch encoder reduces the connection between image style and secret information.The method introduces a mutual learning training method to make dual-branch encoders and decoders with different training directions work together.The image steganography method based on the dual-branch generative adversarial network is proposed.The specific research contents mainly include:Aiming at the security problem of image steganography,a generative image steganography method based on attention mechanism is proposed.The attention mechanism can assign weights to image data in different regions.The attention mechanism can significantly improve the quality of images generated by neural networks and has been successfully applied in the field of convolutional neural networks.Using a generative adversarial network to coverless image steganography,the quality of the generated secret images is directly related to the hiding effect of secret information.The method combines the attention mechanism with the generative adversarial network to make the quality of steganographic images better and the security of secret information better.The method mainly includes the design of the symmetric encryption algorithm and steganographic network.The symmetric encryption algorithm encrypts information symmetrically.The steganography network combines the attention mechanism and the generative adversarial network to make the generated image details clear and rich in texture.Experimental results show that the proposed method is effective.To solve the concealment problem of image steganography,an image steganography based on double-branch generative adversarial network is proposed.This method mainly includes information mapping method and encoder design method.Information mapping method maps secrets into information vectors according to a rule.The encoder is designed with a double-branch structure to hide secrets.Aiming at the accuracy of image steganography,the mutual learning training method is applied to further improve the accuracy of steganography.Experimental results show that the proposed method is effective.