Abstract: A geometrically robust blind watermarking algorithm was proposed using translation, rotation and scaling invariant property of relative coordinate of the invariant centroid. With certain level wavelet decomposition according to the length of the watermark bits, after modifying some selected coefficients in the lowest frequency of the last level for its robustness to the interference, the modified coefficients were applied to obtain the watermarked image by wavelet reconstruction. With the improved invariant centroid based characteristic matrix and the Rijndael encryption,a related binary logical table was produced. As the secret key, the table was used to apply to the third party for copyright proteclion. This arithmetic can extract the watermarking without the original carrier image and the original watermarking. Experimental results show its low time complexity and robustness to the common attacks such as JPEG compression, cropping and added noise, as well as the geometric attacks such as translation, rotation and scaling.

Key words: Invariant centroid, Geometric attacks, Wavcet decomposition, Invariant characteristic matrix, Rijndael encryption