Coverless Information Hiding Based On Deep Generate Model

Data hiding technology is embedding information into image,audio or video content.It not only protects the content of secret files,but also hides the communication process itself,avoiding attackers' attention.Most of the current information hiding technology is to make secret information embedded in the carrier and disguise the secret information by modifying the carrier data(digital image,video and audio).However,the pace of development has slowed in recent years.The detection technology of hidden information(steganography analysis technology)has also developed rapidly.The technology is to judge the existence of secret information through the statistical anomaly of carrier data,and it is a serious threat to information hiding.Therefore,the study of image based steganography and steganography is also more concerned.Because the information hiding is to embed secret information into the carrier after transmission,or the synthesis of secret information and texture,after the transfer,these methods are also carried out the secret information transmission,to achieve information hiding only by transmitting a free secret information carrier,is a challenging problem.The generation model can capture the high order correlation of the observed or visible data without the target class label information,and the trained model can generate new data that conforms to the sample distribution by sampling the effective sample prior knowledge from the network.The input data can automatically discover the internal characteristics of input data,and then find the feature difference of data.Therefore,this dissertation uses the generation model to hide the image information,and uses the Wasserstein generation in the generation model to train the generation model database of the image information hiding.Secondly,this dissertation uses the generated model database of the image information hiding to hide the information of the secret image which needs to be transmitted.The main contents of this dissertation are summarized as follows:An information hiding strategy based on Wasserstein generation antagonism network is proposed for image steganalysis.Because the image steganalysis is through statistical data carrier exceptions to determine the existence of secret information,so we use Wasserstein to generate the generator against network can generate infinite as much as possible and the real data distribution as the data distribution,does not require any additional conditions,through the internal competition mechanism and separatealternating iterative training generator,then constructs the new model,the original image generated by the transmission model of camouflage images and the corresponding can generate the desired transmission,to achieve the same effect with the original image transmission.Put forward a hidden strategy generation model database based on image analysis and information hidden in the third chapter,in the practical application we find that mentioned in the third chapter of the strategy we have to generate model parameters to camouflage image transmission and the corresponding achieve our desired effect,which relates to our model parameters transmission the size of the image,our model parameters transmission than the image dimension,which leads to while we present a novel strategy did not save the channel size.So we send both the sender and receiver to train the model parameters at the same time,and build the new generation model database,so that we no longer need to transfer the model parameters,and only need to transmit images independent of secret images,which will achieve the same effect as the transmission of secret images.