The Research Of Digital Watermark Algorithm Based On SIFT

As the Internet, communication technology and capacity of hardware storage develop rapidly, the protection of digital media copyright becomes more and more important. As a type of significant method for copyright identification and protection, digital watermark caught the attention of information security researchers.The thesis proposes an algorithm which focuses on resistance of geometric attacks, common filtering and compressing attacks. The algorithm which is based on SIFT area descriptor was proposed in 2004. Combined with discrete cosine transform of local image, in which the embedding and extracting of watermark are implemented, the method obtained robust experimental results.The thesis firstly reviews the basic concept and model of watermark, and classifies watermark into two types: Global Watermark and Local Watermark. By analyzing the two types of watermark, the thesis identifies the advantages of local watermark against geometric attacks and other ruins of images. To accomplish robust local watermark, then, the thesis introduces SIFT feature descriptor which is scale-invariant and rotation-invariant.SIFT feature descriptor gets key-points via multi-scale analysis and uses the statistic orientation of neighborhood of key-points to assign main orientation, which is rotation-invariant for the key-points. With the invariant performance of SIFT descriptors, the algorithm calculates DCT coefficients in 9×9 neighborhood of key-points and chooses a part of coefficients in middle frequency domain to proceed watermark embedding. In the process of extracting, with the invariant performance of SIFT descriptors, Radon transform is used for geometric revise of attacked images. The last step is to extract watermark from the DCT coefficients.As demonstrated by the experiments carried out, the method of the thesis which combined watermark with area features of local image is a kind of robust watermark scheme compared with traditional watermark.