Research Of Multimodelity Steganography Based On Generative Adversarial Networks

The steganography is a science that hides secret information in natural carriers without changing the carrier's perception characteristics.Images,videos,voices,texts and other digital media can be used as carriers for steganography,and Audio steganography is a technology that uses audio as a carrier for steganography.The existing audio steganography methods are mainly manual design methods,and a large amount of corresponding professional knowledge is required in algorithm design,the threshold is high,and many challenges are still faced.In addition,in the era of big data,media forms are rich and diverse.If the information hiding model can only be embedded and extracted for a single type of carrier,its security is not enough.We hope that no matter the input carrier object is audio,image or other types of digital media,the information hiding model can embed and extract it,that is,multimodelity steganography technology.Generative adversarial network is a kind of deep unsupervised learning architecture,which can make the Jensen-Shannon divergence between the generated sample distribution and the real sample distribution decrease continuously,that is,to make the generated sample is as close as possible to the real sample.In this paper,digital audio and digital images are taken as the research objects,and the multi-type carrier steganography integrated system based on convolutional neural network and generative countermeasure network is studied.The main research results of the paper are follows:(1)The visual audio steganography model based on convolutional neural network is proposed.The model consists of an encoder and a decoder:The encoder can embed the secret image into the audio carrier,and the decoder can extract the secret image from the secret carrier.This paper proposes to send audio to the convolutional neural network in a visual way to solve the problem that the convolutional neural network can only process images.The proposed scheme is trained and verified on two training sets.The experiment and results show that this scheme can embed pictures in audio and recover secret images with high quality.(2)An integrated steganography system based on generative adversarial network is proposed,which can steganize audio and image simultaneously.This work refers to the existing image steganography model based on generative adversarial network,and combined with the visual audio steganography model based on convolutional neural network,designs an integrated steganography system which can take audio and image as carriers at the same time.The system consists of an encoder,a decoder and a discriminator:the encoder is responsible for embedding the secret image into the audio carrier or image carrier,the decoder is responsible for recovering the secret image from the encrypted audio or image,the discriminator is responsible for judging whether the input sample is a real sample or a generated sample.Experiments show that the system performs well on both audio and image carriers.