Abstract:

A novel self-embedding perfectly blind watermarking algorithm is proposed based on QR
matrix decomposition for copyright protection of digital color images. At first, G channel component of
the original image is performed with non-subsampled shearlet transform, then the low-frequency
component is performed with block-QR decomposition, the feature watermark is derived by judging the
number relationship between vector l1 norm from the first row elements of each sub-blocks R-matrix
and the mean of vector l1 norms from the first row elements of all sub-blocks R-matrix. Secondly, the
low-frequency component is obtained after DWT from B channel of original image, then it is performed
with block-QR decomposition. The feature watermark is embed by modifying the first row and the least
column element of the R-matrix from every blocks. The generation and embedding of feature
watermark are done in two channels independently, the watermark can be detected without the original
image, and the algorithm can identify an image without the additional watermark information.
Experimental results show that the proposed algorithm has strong robustness to resist various common
attacks such as adding noise, JPEG compression, scaling, cropping and row shifting.

Key words: digital watermarking, self-embedding, perfectly blind detection, QR decomposition;
shearlet transform