Graph spectral domain blind watermarking

This paper proposes the first ever graph spectral domain blind watermarking algorithm. We explore the recently developed graph signal processing for spread-spectrum watermarking to authenticate the data recorded on non-Cartesian grids, such as sensor data, 3D point clouds, Lidar scans and mesh data. The choice of coefficients for embedding the watermark is driven by the model for minimisation embedding distortion and the robustness model. The distortion minimisation model is proposed to reduce the watermarking distortion by establishing the relationship between the error distortion using mean square error and the selected Graph Fourier coefficients to embed the watermark. The robustness model is proposed to improve the watermarking robustness against the attacks by establishing the relationship between the watermark extraction and the effect of the attacks, namely, additive noise and nodes data deletion. The proposed models were verified by the experimental results.