NSCT-GBT-SVD Combined with Feature Region Robust Watermarking Algorithm

‍ ‍In view of the poor resistance of most watermarking algorithms to geometry attacks （rotation， translation， scaling， and shearing）， this paper proposed a Non-Subsampled Contourlet Transform（NSCT）-Graph-based transform（GBT）-Singular Value Decomposition （SVD） combined with （multi-scale multi-direction structure tensor-fuzzy C-means clustering） feature region robust watermarking algorithm. Firstly， NSCT transform was used to obtain the low-frequency region， and then the multi-scale and multi-direction structure tensor was established to find the stable feature points. The feature points were divided into three clusters by fuzzy C-means clustering， and the stable and non-overlapping feature regions in each cluster were constructed. Then the watermark embedding region was constructed by the local normalization operation of the image， and the watermark embedding region was divided into blocks and the stable coefficient matrix was obtained by GBT of the block region. Finally， SVD was used to embed the watermark information into the coefficient matrix. Simulation experiments show that the proposed algorithm has good robustness to both conventional signal attacks and geometric attacks when the peak signal-to-noise ratio is higher than 44 dB. Especially， when facing the geometric attacks of rotation， translation， cutting， scaling and pairwise combination， the extracted watermark normalized correlation coefficients are all above 0.94.