A robust watermarking scheme for 3D point cloud models using self-similarity partition

Abstract

This paper presents a robust spatial watermarking scheme for 3D point cloud models using self-similarity partition. In the new scheme, 3D point cloud model is uniquely partitioned into patches using octree structure and PCA preprocessing. Then a concise similarity measurement is designed to identify and cluster similar patches to similar patch chains and the codebook is formed. By altering local vector direction of certain points of each patch, the watermark bits are embedded into the average local vector direction of every similar patch. The watermark can be extracted based on three keys, the known model center, the principal object axis of the original model, and the codebook. Furthermore, to make the proposed watermarking scheme robust against various forms of attack while preserving the visual quality of the models our scheme repeatedly embeds a bit of the watermark in every self-similarity clustered patches. Experimental results show that the proposed watermarking scheme has good imperceptibility and performs well under common 3D watermarking attacks such as uniform affine transformations, simplification, resample, cropping, additive noise.