Coverless Information Hiding Based on Generative adversarial networks

Abstract—Traditional image steganography modifies the content of the image more or less, it is hard to resist the detection of image steganalysis tools. To address this problem, a novel method named generative coverless information hiding method based on generative adversarial networks is proposed in this paper. The main idea of the method is that the class label of generative adversarial networks is replaced with the secret information as a driver to generate hidden image directly, and then extract the secret information from the hidden image through the discriminator. It’s the first time that the coverless information hiding is achieved by generative adversarial networks. Compared with the traditional image steganography, this method does not modify the content of the original image. therefore, this method can resist image steganalysis tools effectively. In terms of steganographic capacity, anti-steganalysis, safety and reliability, the experimen shows that this hidden algorithm performs well
Index Terms—Information security, Coverless Information hiding, Generative adversarial networks.

1 Introduction In the way of traditional steganography, the secret information is embedded into the carriers which is used as the cover of secret information by an invisible way, so as to achieve the purpose of secret communication. The carriers include common digital images, audios, videos and so on [1]. Mostly, traditional steganography directly modifies the carrier to embed secret data. Owing to the partial distortion of the carrier, the third party can detect the existence of hidden secret information by finding the statistic evidence introduced by the embedding method.
Among all the carriers of information hiding, digital images are used most widely. In traditional image steganography, pixel values are modified to achieve the embedding of secret information. According to the different ways of hiding, the common steganography methods can be classified into two categories: hiding methods in spatial domain and transform domain. Steganography in the spatial domain, such as the method proposed in paper[2] which replace the LSB (least significant bits) of the image with secret data, the adaptive LSB hiding method[3], the spatial adaptive steganography algorithm S-UNIWARD [4],HUGO[5],WOW[6] and so on; The transform domain method is to modify the host image data to change some statistical features to achieve data hiding, such as the hidden method in DFT(discrete Fourier transform) domain [7], DCT (discrete cosine transform) domain [8], and DWT (discrete wavelet transform) domain [9].
These methods modify the carrier images to embed the secret information by the certain rules, it is inevitable to leave some traces of modification on the carrier. Hence, we are facing such a problem: these steganography methods cannot resist the detection of existing steganalysis tools. In order to resist the detection of all kinds of steganalysis algorithms fundamentally, researchers have proposed the concept of coverless information hiding[10]. Compareing with traditional information hiding methods, Coverless information hiding does not require extra carriers , but generates or obtains the digital images driven by the secret data directly. Zhou Zhili et al [10] proposed a coverless information hiding based on Bag-of-Words model of image, his idea is to establish the mapping relationship between the original image and the secret information. But This method only avoids the change of embedded information to the image carrier, and the capacities of this method are small while the image database is very large. To solve this problem, we propose a novel method—-generative coverless information hiding method—-which is based on Generative adversarial networks in this paper. In the recent research on Generative Adversarial Networks (GANs) [11] the generation of image samples is driven by noise directly, which is in good agreement with the idea of coverless information hiding. Therefore, we propose a new generative coverless information hiding method: First, the text is encoded and then as a driver, it combined with the encoded information and the noise to generate image samples. The generated image samples are the hidden image for transmissionin order to realize the generative coverless information hiding. Because the carrier image is not modified, the third party will not easily perceive anomalies about stego images. Furthermore, this method can resist detection of all the existing steganalysis tools. 2 Generative adversarial networks GAN is a very efficient generation model proposed by Goodfellow et al [11] in 2014. The structure of GAN is shown in Figure 1. The idea of GAN comes from the two-

…(Full text truncated)…