Digital Watermark Blind Extraction Algorithm Based On Nonsubsampled Contourlet Transform

Digital watermarking technology, as an important research direction in the world, plays a significant role in many fields such as information security and copyright protection. There are some disadvantages that the embedded process or key need to be known in advance for traditional watermarking extraction methods. The less prior knowledge is known, the greater difficulty of extraction is. In view of Discrete Cosine Transform (DCT) watermarking, this paper focuses on blind watermarking extraction algorithms based on a single channel blind separation idea. On the premise of without prior knowledge and according to the statistical independence hypothesis of sources, three kinds of blind watermarking extraction algorithms are respectively proposed in this paper.The main contributions of this paper are as follows:1) A blind extraction algorithm of redundant observations based on nonsubsampled contourlet transform (NSCT) is proposed. A group of sub-band components are obtained by the embedded watermark image through NSCT. For the purpose of adding virtual source signals, the redundant observations are constructed. Redundant source signals are separated by independent component analysis (ICA) algorithm, which can reduce or eliminate irrelevant information between mutual sources. The original and watermark image are estimated through weighted average. Experiments show that the proposed algorithm has a good extraction effect.2) Two blind extraction algorithms of reconstruction observations based on contourlet transform and nonsubsampled contourlet transform are proposed. Subband components, which are respectively obtained by contourlet transform and nonsubsampled contourlet transform, are exchanged or changed based on principle of minimum energy. The new observations are reconstructed by changed subband components. Then the original and watermark image are separated by ICA algorithm. Experiments indicate that the algorithms have effective consequence under different embedded intensities.3) A blind extraction algorithm of redundant observations based on nonsubsampled contourlet transform and principle component analysis (PCA) is proposed. The number of subband components can be reduced by dimension reduction of PCA, which should meet cumulative contribution rate. Reserved components and embedded watermark image are as new observations. Watermark image is extracted by ICA. Experiments demonstrate that the proposed algorithm has efficient anti-attack capability.