Research On Robust Image Steganography Based On Double Generative Adversarial Network

With the rapid development of Internet technology,image information security transmission and digital media copyright protection are becoming more and more serious.Image steganography has always been the focus of information security research,in the secret transmission of information,copyright protection and other important applications.However,at present,image steganography focuses on improving the concealment of the model,but the research on its robustness is relatively lagging behind.This paper first introduces the basic concepts of image steganalysis and related model,the technique of image steganalysis lack of robustness research problems,then raises Deep Convolutional Neural Network(DCNN)and Generative Adversarial Networks image steganographic model for improvement of the normal resolution image steganographic robustness,And scale-invariant feature transform(SIFT)algorithm combined with regional image steganography to improve the steganographic robustness of high-resolution images.The specific content is as follows:(1)An image steganography model based on double generative adversarial networks is proposed.the model is composed of two series generative adversarial networks,which can hide grayscale images into color or grayscale images of the same size and restore them.By enhancing the data of the generated dense images and further strengthening the training of the extraction network,the extraction network can adapt to the geometric transformation of the input images,which can not only realize the steganography of the image information with high capacity,but also resist the geometric attack within a certain range,and has better robustness than the same kind of models.(2)For high-resolution cover images,SIFT feature descriptors with scale invariance,rotation invariance and translation invariance were used to select feature points and determine the local steganography area,and then the steganography model of deep learning was used to embed secret images.During the extraction process,SIFT feature description sub-algorithm is also used to extract secret images after geometric correction of the attacked images.It isproved by a series of experiments that this method is a steganographic model with high robustness.