Abstract:In deep steganography, the model size is usually related to the underlying mesh resolution, and a separate neural network needs to be trained as a message extractor. In this paper, we propose a generative image steganography based on point cloud representation, which represents image data as a point cloud, learns the distribution of the point cloud data, and represents it in the form of a continuous function. This method breaks through the limitation of the image resolution, and can generate images with arbitrary resolution according to the actual need, and omits the need for explicit data for image steganography. At the same time, using a fixed point cloud extractor transfers the training of the network to the point cloud data, which saves the training time and avoids the risk of exposing the steganography behavior caused by the transmission of the message extractor. Experiments prove that the steganographic images generated by the scheme have very high image quality and the accuracy of message extraction reaches more than 99%.

What problem does this paper attempt to address?

This paper attempts to solve the following three main problems: 1. **Limitations of image resolution**: - In traditional steganography schemes, the carrier image is stored in the form of a grid, that is, represented by pixels. When continuous image signals are sampled as discrete grid data, detailed information will be lost, resulting in a decrease in image resolution and quality. Moreover, the complexity of the model is closely related to a specific resolution, making it impossible for a single model to adaptively represent images of different resolutions. - The method proposed in this paper breaks through the limitations of image resolution by representing image data as a point cloud, learning the distribution of point cloud data, and representing it as a continuous function. This method can generate images of any resolution according to actual needs. 2. **Transmission risks and training costs of message extractors**: - In traditional steganography, the transmission behavior of message extractors is likely to arouse suspicion, increasing the risk of exposing steganographic behavior. At the same time, the size of the extractor is usually large, making it difficult to transmit, and the training cost is high. - This paper adopts a fixed point - cloud extractor, shifting the focus of network training to optimizing the point - cloud data itself. This not only reduces the training cost but also improves the quality of steganographic images and avoids the exposure risk caused by transmitting the extractor. 3. **Dependence on explicit image carriers**: - Current image steganography schemes usually rely on explicit image carriers for message embedding and extraction. Even when using implicit neural representations, explicit image sampling is still ultimately required. - The generative image steganography based on point - cloud representation proposed in this paper operates directly on point - cloud data, eliminating the need for explicit image carriers and broadening the range of potential steganographic carriers. ### Overview of solutions To overcome the above problems, this paper proposes a generative image steganography based on point - cloud representation. Specifically: - **Neural implicit representation for image representation**: Learn the function distribution of the image through neural implicit representation, sample functions from this distribution, and use coordinates as input sample features to form point - cloud data to represent the image. This representation method breaks through the resolution limitation, and images of different resolutions can be sampled from the model simply by changing the resolution parameter. - **Optimized point - cloud extractor enhances steganography**: Introduce a fixed point - cloud extractor, shifting the focus of network training to optimizing the point - cloud data itself. This not only reduces the training cost but also significantly enhances the resistance to steganalysis and improves overall security. - **Universal multimedia data steganography framework**: The steganography process acts directly on point - cloud data, eliminating the need for explicit carriers (such as images), establishing a universal steganography framework not limited to image data, and is expected to be extended to other multimedia formats in the future. Through these improvements, this paper provides a new steganography implementation framework and solves several key problems in existing methods.

Generative Image Steganography Based on Point Cloud

Image steganography based on generative implicit neural representation

Generative adversarial networks-based image steganography with multiscale features integration

A Novel Grayscale Image Steganography via Generative Adversarial Network

Batch Steganography Via Generative Network

StegaDDPM: Generative Image Steganography based on Denoising Diffusion Probabilistic Model

Provably Secure Generative Steganography Based on Autoregressive Model

Generative Steganography Network

Pixel-Stega: Generative Image Steganography Based on Autoregressive Models

Establishing Robust Generative Image Steganography Via Popular Stable Diffusion

Channel Attention Image Steganography With Generative Adversarial Networks

Generative Image Steganography based on Guidance Feature Distribution

Diffusion-Stego: Training-free Diffusion Generative Steganography via Message Projection

Cover-separable Fixed Neural Network Steganography via Deep Generative Models

Spatial Image Steganography Based on Generative Adversarial Network

LDStega: Practical and Robust Generative Image Steganography Based on Latent Diffusion Models

HidingGAN: High Capacity Information Hiding with Generative Adversarial Network.

Lossless image steganography: Regard steganography as super-resolution

Provably Secure Robust Image Steganography

High invisibility image steganography with wavelet transform and generative adversarial network

Sstegan: Self-Learning Steganography Based On Generative Adversarial Networks