An Adaptive Steganography for 3D Point Cloud Models

This paper presents a new adaptive, high-capacity steganography for 3D point cloud models using self-similarity segmentation. Every embedding vertex of the model can adaptively embed variable bits using an adaptive space subdivision procedure which uses normal direction of vertexes to estimate the embedding capacity of every vertex with respect to human visual system. The scheme segments the 3D point cloud model to patches using self-similarity measures. The representative patches and similar patches are then taken as the reference patches and the message patches. Finally, every message point in the similar message patches which has the point-to-point correspondence with a reference point in the reference patch can adaptively embed at least three bits by shifting the message point from current point to the corresponding embedding position using space subdivision. Experimental results show that the proposed scheme is adaptive, has high capacity and low distortion, and is robust against uniform affine transformations such as transformation, rotation, scaling.