Vector Polygon Blind Watermarking Based on Canonical Correlation Analysis

A vector polygon blind watermarking algorithm based on canonical correlation analysis (CCA) is proposed for protecting the copyright of vector polygon. In embedding, vertices of a polygon are separated into two groups according to odd-even properties of their indexes. CCA is carried out on the two vertex groups to get canonical variables. The absolute value of canonical variables is proved to be affine invariant so that it is called affine invariant descriptor. The watermark is embedded into the magnitude of Fourier descriptors of affine invariant descriptors. In detection, affine invariant descriptors are obtained through a same process as embedding. The correlation between magnitude of the Fourier descriptors of affine invariant descriptors and the watermark is calculated. The watermark can be detected by comparing the correlation to a pre-selected threshold. The algorithm is blind as it doesn´t need the original graphics during detection. Both theoretical analysis and simulation results have demonstrated feasibility and reliability of the proposed algorithm.