Research On Steganography Of High-capacity Images Based On Deep Convolutional Neural Network

Images are the most important form of information expression among multimedia tools,and are widely used in many fields such as business management,education and teaching,military training,and family life.The development of the Internet and multimedia technologies has resulted in the storage and transmission of large amounts of image data on the network,and a large part of these image data involves secret information such as personal privacy or business secrets.Therefore,security of image data transmission and storage is urgently needed to solve.Image steganography is a technique that hides secret images into non-secret carriers,thus hiding the existence of secret images and achieving secure transmission of information.Traditional image steganography generally embeds and extracts data through a manually designed cost function.Steganography has low efficiency and cannot be dynamically adjusted in real time for third-party detection tools.Therefore,traditional image steganography is gradually falling into the bottleneck.In recent years,the deep neural network model has made a series of breakthroughs in many fields such as image processing,natural language processing and speech recognition.It has also gradually penetrated in the field of information steganography and has shown great potential.The current main research on image steganography has the following two contents: one is to ensure the visual quality of non-secret carriers under the premise of improving the steganography capacity of the information;the other is to improve the security of steganography itself.Therefore,for the above two problems,this paper proposes a high-capacity image steganography method based on deep convolutional neural network.The main work includes the following two points:(1)In order to improve the image steganography capacity and steganography efficiency,we propose an image steganography method based on SegNet structure.The network structure includes two sub-networks: hiding network and revealed network,and the two networks are trained simultaneously.First,train the two sub-networks in advance;second,embed the entire natural image(secret image)into the non-secret carrier image through the hiding network at the sender,and finally obtain the secret image;finally,the receiver will encrypt the image The image extracts the secret-related image through the revealed network.Through experimental simulation and analysis,the network model can achieve steganography of full-size images,and can ensure the visual quality of the generated images,and the deep neural network can batch process the data,and the steganography efficiency has been improved.(2)In order to further improve steganography security and information payload capacity,we added ECC and VQ-VAE on the basis of SegNet network.First,the non-secret image is compressed and reconstructed through the VQ-VAE network,and then ECC encryption is performed,and finally the encrypted image is obtained.Finally,the encrypted image is steganographed and extracted through the hiding network and the revealed network in the SegNet network.Through experimental simulation and analysis,after adding VQVAE,its information payload has been enhanced.ECC is an asymmetric encryption technology with short keys and high security,so the steganography framework has double security guarantees.