Reserach On Coverless Informatiom Hiding Algorithm Based On Generative Adversarial Networks

In recent years,with the widespread use of mobile Internet devices,the Internet has penetrated into every aspect of people’s daily life,which has attracted people’s attention to information security.Because of the privacy of information hiding,it plays an irreplaceable role in dealing with information security issues.The embedding-based steganography embeds secret information into the cover image according to the specified rules,which destroys the statistical features of the cover image and makes it difficult to resist the detection of steganalysis algorithms.With the development of deep learning,generative adversarial networks(GAN)have shown significant results in the fields of image generation and processing.The coverless information hiding algorithm based on generative adversarial network directly drives the generative model to generate stego images through secret information.There is no embedding operation,which ensures the security of the stego image during the transmission,and shows important research significance.However,the current development of the coverless information hiding technology based on generative adversarial networks is still in its infancy,and the design of related hiding algorithms is not perfect.After a deep understanding of the current research status,we propose two coverless information hiding algorithms based on generative adversarial networks.The main work is as follows:1)A noise-robust coverless information hiding algorithm based on generative adversarial networks is proposed.The stego image will inevitably be interfered by noise during the transmission.On the basis of the existing hidden algorithm,the denoising network is added to reduce the gap of the stego image between the sender and the receiver.At the same time,during the process of model training,the stego image is randomly attacked by using noise to simulate the transmission environment of the stego image in the real channel.In the stage of secret information recovery,by designing a noise vector extractor,the secret accuracy of information recovery is 97%.Compared with the same type of algorithm,this method has better noise robustness on the basis of large information hiding capacity.2)An information hiding strategy based on image attribute editing is proposed.Firstly,mapping the secret information to image attribute labels according to the designed mapping table,and driving the image attribute editing model Att GAN together with the face images generated by generative adversarial network to generate stego images.In the recovery stage of secret information,the corresponding attributes are extracted by attribute label classifier.Thereby,the secret information is recovered.The stego image generated by this method is only a simple modification of the relevant attributes of the face images generated by generative adversarial network,and shows good visual quality.Therefore,there is better concealment during the transmission,reducing the possibility of third-party attacks.In addition,this strategy does not need to share the image database between the sender and receiver,which saves storage resources.Compared with the same type of algorithm,the designed attribute classifier has better accuracy of classification,thus a higher accuracy of secret information recovery is achieved.