Extracting embedded data in 3D models from 2D views using perspective invariants

Abstract

A 3D-2D watermarking method using perspective projective invariance is proposed. Data is embedded in relative positions of six points on a 3D mesh by translating one of them, and extracted from any 2D view generated as long as the points remain visible. To evaluate the performance of the perspective invariant, a watermarking system with a very simple interest point detection method is implemented. Simulations are made on six 3D meshes with different watermark strengths and view angles. Very promising results show that the used perspective invariant is suitable for 3D-2D watermarking, opening a completely new area of research.