基于边界平衡生成对抗网络的生成式隐写

Generative Steganography Based on Boundary Equilibrium Generative Adversarial Network

针对基于深度生成模型的载体合成隐写中生成含密图像质量不高的问题,提出了一种基于边界平衡生成对抗网络(boundary equilibrium generative adversarial network,BEGAN)的生成式隐写方法。该方法首先将秘密信息分段并转化为噪声,而后将噪声输入到BEGAN生成器中得到高质量的含密图像;在消息提取阶段,从含密图像输入噪声提取器中恢复出噪声,最后将噪声转化为秘密信息并完成消息提取。首次提出了针对含密图像的质量评价问题。实验结果和分析表明,与基于深卷积生成对抗网络(deep convolution generative adversarial networks,DCGAN)的隐写方法相比,采用基于编码-解码结构的BEGAN模型能够提高含密图像质量,增大输入噪声的长度能够提高隐写容量。

Focused on the issue that the low quality of the steganographic images generated in the cover synthesis steganography based on deep generative model,a steganographic method based on the boundary equilibrium generative adversarial network(BEGAN)was proposed.Firstly,the secret information was segmented and converted into noises.Then,the noises were input into the BEGAN generator to get high-quality stego-images.In the information extraction stage,the noises were recovered from the noise-extractor with the stego-image.Finally,the noises were converted into secret information to complete the information extraction.Experimental results and analysis showed that compared with the DCGAN-based steganographic method,the BEGAN model based on the encoder-decoder structure could improve the quality of stego-images,and increased the input noise length could improve the steganographic capacity.