A robust image watermarking using spatial-frequency feature of wavelet transform

We propose a robust watermarking scheme for digital images using the spatial-frequency feature of wavelet transform. For a wavelet coefficient array, we extract a completed tree from a coefficient in the LL subband to construct a full-tree matrix (FTM), which represents both of the spatial localizability and multiresolution decomposition of an overlapped local region in an original image. Based on the FTM, three different features, i.e. the local spatial luminance sensitivity, the frequency sensitivity and the texture activity, can be exploited to determine whether the watermark is inserted to this FTM or not. Experimental results show that the watermarking sequence is inserted in those regions that are of high texture activity with the maximum strength of JND tolerance. It is also shown that better compromise between the robustness and transparency is obtained using our proposed scheme