The Research Of Digital Watermark Embedding Algorithm Based On The Transform Domain

Digital watermark is a new and effective way to protect digital products and to maintain the data security. It is a very practical information hiding technique. Watermarking utilize a digital method to embed some markers which have special meanings into digital images, audio, documents, books and video. Watermarks can testify the ownership of the above. What is more, watermarks can also be the evidence in front of the court. In a word, digital watermarks can be an effective method to protect knowledge property and multimedia data. However, watermarks should not destroy the usefulness and the value of being viewed of the original data. Now any companies have developed digital ideograph systems and put them into operation in electronic business, digital works dealing systems,video processing systems. Digital watermarks have also find its usage in military secret communication and communication counterwork. Watermarks are used in copyright protection, and its potential use lies in electronic business, online or off-line distribution of multimedia data and mass digital broadcasting. Watermarks are used to hide markers, and its applications include medical examination, drawing, digital imaging, digital video watch, multimedia index and content-based searching.Two digital watermark embedding algorithms in transform domain are presented, which are based on analysis and research of DCT, DWT and Neural Network. With the coefficients after Discrete Cosine Transform (DWT) of the digital image having correlations, we firstly consider making another Discrete Cosine Transform (DCT) of them, and then embedding digital watermark. The algorithm has the characteristics of both DWT and DCT, so enhances the robustness. The other algorithm which is embedding a binary digital watermark into a discrete wavelet transformed image combining neural network is proposed: Firstly, after discrete wavelet transform of the original image, we take the low-frequency coefficients as embedding location, select some pixels and their vicinities randomly among the low-frequency coefficients. Then we establish the neural network model, and train to get the output value of the network. Finally we embed watermark information by adjusting the polarity between a selected pixel and the output value of the network. The algorithm is a non-blind algorithm, so in the extracting process we have no need to use the original image and watermark. In both of the above algorithms, to enhance the security and against geometry attack of the ability, the digital watermark is pre-treated by Arnold encryption. Binary image and gray image are used as the watermark signal in this dissertation. Comparing with traditional watermark, they include more information of copyright ownership, and these watermarks after being extracted are convenient to observe directly. Finally, in order to verify the imperceptibility and robustness of the algorithms proposed in this paper, a lot of experiments have been done. The experimental results show that both of the algorithms have good imperceptibility and very strong ability against geometry attack and common image processing operations. At last a conclusion and prospect of this paper have been done.