Abstract:Steganography is a collection of techniques for concealing the existence of information by embedding it within a cover. With the development of deep learning, some novel steganography methods have appeared based on the autoencoder or generative adversarial networks. While the deep learning based steganography methods have the advantages of automatic generation and capacity, the security of the algorithm needs to improve. In this paper, we take advantage of the linear behavior of deep learning networks in higher space and propose a novel steganography scheme which enhances the security by adversarial example. The system is trained with different training settings on two datasets. The experiment results show that the proposed scheme could escape from deep learning steganalyzer detection. Besides, the produced stego could extract secret image with less distortion.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the security of steganography methods based on deep learning. Specifically, the author points out that although the existing steganography methods based on Generative Adversarial Networks (GAN) have advantages in automatic generation and capacity, their security is poor when facing independent steganalysis tools. To enhance security, this paper proposes a new steganography scheme, which improves the ability of steganographic data to evade detection by using adversarial example techniques. This method can not only make the generated stego - images escape the detection of deep - learning steganalysis tools, but also cause less distortion to the extracted secret images. ### Main contributions: 1. **Proposed a new steganography scheme**: Combined GAN and adversarial example techniques to enhance the security of stego - images. 2. **Experimental verification**: Through experiments on two datasets, the effectiveness of the proposed method was proved, especially when facing independent steganalysis tools. 3. **Application of adversarial example techniques**: Two adversarial example techniques, Fast Gradient Sign Method (FGSM) and One Pixel Attack, were used to test white - box and black - box attack scenarios respectively. ### Key technical points: - **Generative Adversarial Networks (GAN)**: Used to generate stego - images and extract secret information. - **Adversarial example techniques**: By adding small perturbations to the generated stego - images, they can escape the detection of steganalysis tools. - **Experimental setup**: Extensive experiments were carried out on two datasets, LFW and Bossbase, to evaluate the performance under different parameter settings. ### Experimental results: - **Model training**: The generated stego - images are very close to the original images visually, and the PSNR value is high, indicating that the generated image quality is good. - **Security enhancement**: By introducing adversarial example techniques, the generated stego - images can effectively escape the detection of independent steganalysis tools, and the detection error rate is significantly increased. ### Conclusion: The method proposed in this paper effectively improves the security of steganography based on deep learning, especially when facing independent steganalysis tools. Future research directions include exploring more general adversarial example techniques and further reducing the impact of adversarial perturbations on secret information.

Enhancing the Security of Deep Learning Steganography via Adversarial Examples

A Novel Grayscale Image Steganography via Generative Adversarial Network

Adversarial Examples Against Deep Neural Network based Steganalysis.

On Security Enhancement of Steganography via Generative Adversarial Image

Spatial Image Steganography Based on Generative Adversarial Network

Seek-and-Hide: Adversarial Steganography Via Deep Reinforcement Learning.

Adversarial Steganography Based on Sparse Cover Enhancement.

Improving image steganography security via ensemble steganalysis and adversarial perturbation minimization

Adversarial feature hybrid framework for steganography with shifted window local loss

Generative adversarial networks-based image steganography with multiscale features integration

Channel Attention Image Steganography With Generative Adversarial Networks

A Novel Technique for Image Steganalysis Based on Separable Convolution and Adversarial Mechanism

Deeply‐Recursive Attention Network for Video Steganography

Direct Adversarial Attack on Stego Sandwiched Between Black Boxes

High invisibility image steganography with wavelet transform and generative adversarial network

CNN Based Adversarial Embedding with Minimum Alteration for Image Steganography

HAG-NET: Hiding Data and Adversarial Attacking with Generative Adversarial Network

A Novel Steganography Method for Character-Level Text Image Based on Adversarial Attacks

Hiding Data in Colors: Secure and Lossless Deep Image Steganography via Conditional Invertible Neural Networks

Sstegan: Self-Learning Steganography Based On Generative Adversarial Networks

Audio Steganography with Speech Recognition System