A Semi-Blind Watermarking Scheme for Color Images Using a Tree Structure

In this paper, we use a tree structure to embed the watermark in DWT decomposition. In the watermark embedding algorithm, the RGB image is first converted to the YUV model. After computing the DWT of the luminance layer, the same PRN sequence is embedded into the DWT coefficients higher than a given threshold T1 in the LL2 and HH2 bands. The PRN sequence is then embedded into the children of the DWT coefficients in the LL2 and HH2 bands. In the final step, the inverse DWT is computed to obtain the watermarked image I. In the watermark detection algorithm, the watermarked RGB (and possibly attacked) image is converted to the YUV model. After computing the DWT of the luminance layer, all the DWT coefficients higher than a given threshold T2 in the LL2 and HH2 bands are selected. The next step is to compute the sum Z, where i runs over all DWT coefficients higher than a given threshold T2 in the LL2 and HH2 bands. This is repeated for the children of modified DWT coefficients in the previous step. In the final step, a predefined threshold T is chosen for LL and HH bands. In each band, if Z exceeds T, the watermark is present. Experimental results indicate that detection in the LL band is robust for one group of attacks, and detection in the HH band is robust for another group of attacks. In future work, we will use this approach to watermark video sequences.