Research On Information Hiding Schemes Based On Generative Adversarial Network

With the rapid development of Internet technology and the diversification of signal transmission channels,the concealment and transmission of sensitive information are facing many risks in this context.Information hiding technology provides an effective means for the transmission of sensitive information,and steganography and digital watermarking are its two different applications.However,with the popularity of information hiding detection devices traditional information hiding techniques are facing great challenges.Therefore,designing an information hiding algorithm with high embedding capacity and low carrier distortion is urgent to solve the secret information leakage and copyright protection.This thesis studied information hiding algorithm to meet this demand.In this thesis,we use generative adversarial networks to design a“map-hiding”information hiding algorithm and present a solution to the problem of carrier image distortion.The details of the thesis are as follows.1.Information hiding algorithm based on generative adversarial network with images as carriersBy constructing a “map-hiding”information hiding model through generative adversarial networks,the embedding and extraction of secret information can share the same parameters,and the embedding and extraction performance can be trained under the same model.Compared with the traditional information hiding techniques combined with deep learning,this method has significantly improved the embedding capacity.2.Solution for the image distortion problemConvolutional neural network has the problem of insufficient extraction in extracting image features,which will lead to distortion of the carrier image in the information hiding model proposed in this thesis will expose the existence of secret information.To solve this problem,in the convolutional neural network to extract the carrier image and secret image feature information introduced fully connected neural network and residual block,the maximum degree of retention of the image feature information.Based on the proposed information hiding model,the embedding capacity reaches3 bpp,the similarity between the carrier image and the secret image reaches 95%,and the accuracy of extracting the secret information even when the secret image is used for the information hiding model.The similarity between the carrier image and the cryptographic image reaches 95%,and the accuracy of extracting secret information reaches more than95% even when the cryptographic image is.The accuracy of extracting secret information reaches more than 95% even in the case of distortion simulation of the carrier image.