| Digital images have become a widely used cover in the field of steganography because of their fast acquisition and processing,abundant redundant information and popular application,etc.Steganography based on digital images has become a research hotspot and focus in the academic community and even industry.In recent years,most of the image steganography methods focus on how to improve the embedding capacity while ignoring their own security.Excessive pursuit of the transmission capacity for secret information makes the embedding behavior easy to be discovered and attacked by the third parties.Therefore,it is of great practical significance to study image steganography with high security.This research is dedicated to improving the security of current image steganography methods.Leading-edge technologies such as encryption technology,Generative Adversarial Network(GAN),Style Transfer,and image de-colorization and colorization are applied to the field of steganography for realizing secure and covert communication of secret information in public channels.In addition,style transfer based image steganography is proposed and a de-stylized network corresponding to the stylized network is designed to achieve the extracted task of secret information.The influence of the embedding operation of secret information on the cover image is successfully eliminated through operations such as style change and color conversion,and the security of the image steganography is correspondingly improved.The research work of this paper are summarized as follows:(1)An Image steganography scheme based on chaos encryption and generative adversarial network is proposed in this paper.The chaotic scrambling and generative adversarial network are applied to the field of image steganography.First,the secret image is chaotically encrypted,and then the encrypted secret image is hidden into the cover image through the embedding network.In addition,the visual quality of the generated steg image is enhanced with the help of generative adversarial network.Compared with deep steganography,the security has been significantly improved.(2)An encrypted coverless image steganography scheme based on generative networks is proposed in this paper.Firstly,the network architecture of Cycle-GAN is borrowed to build the basic framework of the encrypted coverless steganography model.With the help of Cycle-GAN’s weakness—the information about the source image is hidden into the generated image in the form of high-frequency signals,a high-frequency extraction module is designed to improve the visual quality of the generated steg images and the reconstructed public images.In addition,to more objectively evaluate the extraction effect of secret information,the MNIST dataset is pasted on the public image as secret information,and the security and feasibility of the steganography method were verified by calculating the recognition accuracy of the extracted MNIST image.(3)A heterogeneous cover image steganography based on quaternion exponent moments and style transfer is proposed in this paper.which can eliminate the influence of the embedding operation of secret information on the cover image by changing the style of the steg image.For the sender,the stylized operation is regarded as an attack,and a robust embedding algorithm based on quaternion exponential moments is selected to embed the secret information into the cover image.For the receiver,the corresponding de-stylized network is designed to reconstruct the steg image by removing the style of the stylized image,and then the secret information is extracted.Although the steganography scheme is not lossless,since the secret information is image-level,the secret information extracted from the reconstructed steg image can retain its appearance unaffected,which is acceptable for the steganography scheme.(4)An image steganography based on de-colorization and colorization is proposed in this paper.For sender,the cover image is modified twice.The first is the embedding operation of the secret information and the second is to de-colorize the steg image to obtain the grayscale image,which can be transmitted on the public channel.For the receiver,the colorized network is designed to reconstruct the steg image and the secret information can be extracted.Color conversion can effectively eliminate the distortion of embedding operation of secret information on the appearance of cover image.It is difficult for an attacker to determine whether the grayscale image is embedded with secret information,and even if it is exposed,the secret information cannot be extracted. |
