Security-Aware Geometrically Robust 3D Watermarking: Threat Models, Games, and Evaluation Protocols
Year: 2026
Type: conference-paper
Abstract: Geometrically robust 3D watermarking is a practical mechanism for protecting digital graphical assets used in CAD/CAM, entertainment pipelines, digital twins, and additive manufacturing. However, many works still equate “robustness” with survival under routine processing (e.g., smoothing or decimation) and rarely specify what an adaptive adversary can observe, query, or optimize. Building on a review of spectral-geometric invariants under realistic 3D acquisition distortions, this paper proposes a security-aware blueprint for evaluating and designing robust 3D watermarks under explicit threat assumptions. We contribute three methodological layers. First, we formalize a threat model capturing representation changes (mesh ↔ point cloud), acquisition-reconstruction loops, oracle access to embedding or detection services, collusion across many samples, and a utility constraint that prevents trivial “removal by destruction.” Second, we introduce gamebased security definitions for removal, forgery, ambiguity, payload manipulation, and key recovery, with measurable adversarial advantages at a fixed detector operating point (controlled false-alarm rate). Third, we propose a parameterized attack taxonomy and a reproducible attacklibrary interface supporting composed attacks, adaptive score minimization, and standardized metadata logging. The resulting protocol reports both robustness curves (TPR/BER vs. attack strength) and security curves (e.g., key-recovery advantage vs. the number of observed assets). We illustrate the blueprint on spectral-geometric invariant watermarking, highlighting eigenbasis drift under remeshing/reconstruction and key-leakage risks under key reuse.
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7202248489
