Adaptive 3D Mesh Steganography Based on Feature-Preserving Distortion

Yushu Zhang, Jiahao Zhu, Mignfu Xue, Xinpeng Zhang, XiaoChun Cao

3D mesh steganographic algorithms based on geometric modification are vulnerable to 3D steganalyzers. In this paper, we propose a highly adaptive 3D mesh steganography based on feature-preserving distortion (FPD), which guarantees high embedding capacity while effectively resisting 3D steganalysis. Specifically, we first transform vertex coordinates into integers and derive bitplanes from them to construct the embedding domain. To better measure the mesh distortion caused by message embedding, we propose FPD based on the most effective sub-features of the state-of-the-art steganalytic feature set. By improving and minimizing FPD, we can efficiently calculate the optimal vertex-changing distribution and simultaneously preserve mesh features, such as steganalytic and geometric features, to a certain extent. By virtue of the optimal distribution, we adopt the Q-layered syndrome trellis coding (STC) for practical message embedding. However, when Q varies, calculating bit modification probability (BMP) in each layer of Q-layered will be cumbersome. Hence, we contrapuntally design a universal and automatic BMP calculation approach. Extensive experimental results demonstrate that the proposed algorithm outperforms most state-of-the-art 3D mesh steganographic algorithms in terms of resisting 3D steganalysis.

Knowledge Graph



Sign up or login to leave a comment