Research On Digital Image Watermarking Techniques Based On Neural Network

The fast development of Internet makes it convenient to access the multimedia data, but at the same time the Internet has brought a new challenge to the copyright protection. As an effective means for the protection of digital property rights, the technique of digital watermarking is an important part of the region of information hiding and attracts wide attention. Digital watermarking is a technique that embeds copyright information, such as figures, sequence numbers, words, image logos and so on, imperceptibly into multimedia data. It can be used for tracing the distribution and usage of multimedia data. The special application of digital watermarking technique requires that the embedded watermark should be not only transparent to human observers, but also robust enough so that it cannot be easily destroyed or removed after some digital image processing or attacks.The thesis focuses on the applications of neural networks in digital image watermarking algorithms. The basic concepts, algorithmic frames and application areas are described; the basic theory of neural network and the present situation of its applications in digital watermarking are described; some new schemes about image watermarking are presented. The main points of this thesis are as follows:(1) The basic concepts, algorithmic frames and application areas are described; then according to the different ways of information hiding, the existing algorithms of digital image watermarking are concluded and compared. The basic theory of neural network and the present situation of its applications in digital watermarking are described.(2) A new image watermarking algorithm in spatial domain using RBF and Hopfield network is presented. The algorithm has the advantage of transparence, robustness and no original image is needed in extraction process. Experimental results show that the algorithm has stronger robustness in noise adding, cropping and JPEG compression.(3) A new digital image watermarking algorithm based on block singular value decomposition (SVD) and BP network because of their excellence is proposed.The singular values of each image sub-block is gained to form a vector by applying the SVD procedure.Then the singular value of embedded sub-block is amended, BP network built is applied to the watermarking of embedment and extraction. The e algorithm recovers the watermark without any reference to the original image. Experimental results show that the new watermarking method performs well in both imperceptibility and robustness for image processing and attack, especially for geometric distortion.