Deep Multi-image Hiding with Random Key

Wei Zhang, Weixuan Tang, Yuan Rao, Bin Li, Jiwu Huang

Abstract: Multi-image hiding is the technique of hiding multiple secret images within one cover image. In most existing methods, it is possible for one receiver to reveal other receivers’ secret images. To improve the privacy and secrecy among different receivers, one possible solution is to introduce the key mechanism, wherein only the receiver with private key has the permission to reveal the corresponding secret image. In this paper, a multiple image hiding method called DEMIHAK (Deep Multiple Image Hiding with Random Key) is proposed, which utilizes deep neural networks to implement a secure key verification. From the side of the sender, each secret image is assigned with a random key, according to which can sample a key map. Then, BindNet is utilized to incorporate a secret image and its key map into a processed secret image, and HideNet is adopted to conceal multiple processed secret images within cover image and generate a stego image. From the side of the receiver, according to a transmitted private key, RevealNet can be applied to reveal the corresponding secret image from the stego image. Experimental results show that DEMIHAK outperforms existing method from the perspective of visual quality, security, and secrecy.