2D Digital Watermarking Based On The Contourlet Transform

The digital watermark, as a powerful method for intellectual property protection andcertification, becomes a hot research direction in information security filed recently. In thispaper, based on analysis of the existing wavelet transform digital watermarking technology,digital watermarking technology is studied based on contour transform that not only has thecharacteristics of wavelet transform, but also possesses the advantages of more direction andanisotropy. The details are as follows:Digital watermarking algorithm is proposed based on contour transform. Firstly, the hostimage is decomposed using Contourlet transform. Secondly, the low frequency sub-band dataare sorted and the appropriate embedding strength coefficient is selected. Then, the watermarkis embedded into low-frequency coefficients in the case of the invisible. Finally, theinvisibility and robustness of algorithm is verified by experiment.Digital watermarking algorithm is proposed based on nonsubsampled contour transform(NSCT) and support vector machine (SVM).Using anti-noise characteristics of the NSCT,combining with the nature of the global optimization of support vector machine, a new digitalwatermark embedding and extraction algorithm is proposed based on NSCT. Firstly, the hostimage is processed using NSCT.Then, according to the local correlation of image, combiningwith the characteristics of human visual system, each pixel of the image is automaticallypreliminary clustering using fuzzy clustering analysis, and then classified property values isfound. Secondly, the training sample set of support vector machine (SVM) classification areselected from each cluster categories above a certain threshold value of the pixel and supportvector machine classification model is set up. Then, according to this model, each pixel of theimage is classified again, the best embedding watermark position is selected using optimizing.Finally, selecting the appropriate embedding strength coefficient, the watermark is embeddedinto low-frequency coefficients in the case of the invisible. And experiments verify that thealgorithm is not visible, robustness, anti-pan and anti-noise characteristics.