基于生成对抗网络的隐写术设计

Design of steganography based on generative adversarial networks

随着数字多媒体的发展,网络数字媒体已经逐渐成为人们传递和获取信息的主要方式,同时,以数字媒体为载体的隐写术也得到了空前的发展。新型隐写术层出不穷,然而,据统计目前的隐写术在大部分情况下都被非法使用,因此,设计安全的隐写方法迫在眉睫。本文提出一种基于生成对抗网络(GAN)的隐写术,主要包括生成网络和判别网络,生成网络主要生成用于隐写的图像载体,判别网络主要区分原始图像和生成图像以及生成图像和生成图像经过嵌入得到的隐写图像。同时,在CelebA人脸数据集上进行了实验,验证本文提出方法的有效性和鲁棒性。

With the development of digital multimedia the network digital media has gradually become the main way for people to transmit and acquire information.Therefore the steganography based on digital media has also experi enced an unprecedented development.However it is estimated that the current steganography is illegally used in most cases.Therefore the design of secure steganography is imminent.In this paper a novel strategy of steganog raphy based on generative adversarial networks is proposed which includes a generative network and a discrimina tive network.The former mainly generates image carrier for steganography while the latter is utilized to distinguish the original images from generated images as well as the generated images from the stegos obtained by the genera ted images.At the same time experiments are conducted on the CelebA dataset to verify the effectiveness and ro bustness of the proposed method.