Feature-Preserving Tensor Voting Model for Mesh Steganalysis.

Area of Science:

• Computer Vision
• Digital Image Forensics
• Geometric Modeling

Background:

• Standard tensor voting is effective for feature detection in 3D models.
• Existing 3D mesh steganalysis methods lack analysis of neighborhood face correlations, limiting their effectiveness.
• Steganalysis aims to detect hidden data within digital media, including 3D models.

Purpose of the Study:

• To propose a neighborhood-level representation-guided tensor voting model for enhanced 3D mesh steganalysis.
• To reveal subtle artifacts introduced by data embedding in 3D meshes.
• To improve the discrimination accuracy between cover and stego meshes.

Main Methods:

• Utilized a tensor voting model incorporating neighborhood-level representations for 3D mesh steganalysis.
• Performed normal voting tensor (NVT) operations on both original and smoothed mesh faces.
• Extracted features based on eigenvalue differences between original and smoothed tensors, followed by nonlinear mapping.

Main Results:

• The proposed feature sets demonstrated superior performance compared to state-of-the-art methods like LFS64 and ELFS124.
• The method effectively captured intricate relationships among vertices, revealing data embedding artifacts.
• High discrimination accuracy was achieved across various steganographic schemes.

Conclusions:

• The neighborhood-level representation-guided tensor voting model offers a powerful approach for 3D mesh steganalysis.
• The proposed feature extraction method significantly enhances the detection of hidden data in 3D meshes.
• This technique provides a robust solution for identifying stego meshes, outperforming current state-of-the-art methods.