GS-Hider: Hiding Messages into 3D Gaussian Splatting

TL;DR

GS-Hider introduces a novel 3D Gaussian Splatting steganography framework that hides multimodal content inside a 3DGS point cloud without compromising rendering fidelity. It replaces the original SH color representation with a coupled high-dimensional feature $\boldsymbol{f}_i$, renders a coupled feature map $\mathbf{F}_{coup}$, and uses two decoders to separately recover the original scene and the hidden message, with the message decoder kept private. The approach achieves high fidelity, strong security against steganalysis, and large capacity, including hiding an entire 3D scene or a 2D image, and even multiple hidden scenes in a single 3D scene. Empirical results on the Mip-NeRF360 dataset show GS-Hider attains robust performance, real-time rendering (~45 fps), and substantial storage efficiency, making it practical for encrypted transmission and copyright protection in 3D assets.

Abstract

3D Gaussian Splatting (3DGS) has already become the emerging research focus in the fields of 3D scene reconstruction and novel view synthesis. Given that training a 3DGS requires a significant amount of time and computational cost, it is crucial to protect the copyright, integrity, and privacy of such 3D assets. Steganography, as a crucial technique for encrypted transmission and copyright protection, has been extensively studied. However, it still lacks profound exploration targeted at 3DGS. Unlike its predecessor NeRF, 3DGS possesses two distinct features: 1) explicit 3D representation; and 2) real-time rendering speeds. These characteristics result in the 3DGS point cloud files being public and transparent, with each Gaussian point having a clear physical significance. Therefore, ensuring the security and fidelity of the original 3D scene while embedding information into the 3DGS point cloud files is an extremely challenging task. To solve the above-mentioned issue, we first propose a steganography framework for 3DGS, dubbed GS-Hider, which can embed 3D scenes and images into original GS point clouds in an invisible manner and accurately extract the hidden messages. Specifically, we design a coupled secured feature attribute to replace the original 3DGS's spherical harmonics coefficients and then use a scene decoder and a message decoder to disentangle the original RGB scene and the hidden message. Extensive experiments demonstrated that the proposed GS-Hider can effectively conceal multimodal messages without compromising rendering quality and possesses exceptional security, robustness, capacity, and flexibility. Our project is available at: https://xuanyuzhang21.github.io/project/gshider.

Figures (12)

• Figure 1: Application scenario of the proposed GS-Hider. The 3DGS trainer (Alice) requires the training views of the original and hidden scenes to train our GS-Hider, comprising a 3DGS point cloud file, a scene and message decoder. Then, Alice will upload the 3DGS point cloud file and the scene decoder online. 3DGS users (Bob) can render the original 3D scene, while only the trainer is authorized to extract the hidden 3D scene, realizing copyright protection or secret communication.
• Figure 2: Comparison of original 3DGS and two intuitive approaches of 3DGS steganography, namely adding an SH coefficient, and optimizing 3DGS and a message decoder.
• Figure 3: Overview framework of the proposed GS-Hider. It uses a coupled secured feature attribute $\boldsymbol{f}_i$ and the rendering pipeline to fuse hidden and original information, obtaining a rendered coupled feature $\mathbf{F}_{coup}$. Then, the scene and message decoder is adopted to decouple the rendered RGB scenes and hidden messages.
• Figure 4: Comparison visualization results of our proposed GS-Hider and other potential methods. The first row of each group: original scene, the second row of each group: hidden scene.
• Figure 5: ROC curve of different methods under StegExpose. The closer the curve is to the reference, the method is better in security.