The Research About Adaptive Watermark Embedding Algorithms For Static Gray Images In DCT Domain

The main content of this thesis is research about adaptive watermark embedding algorithms for gray images in DCT domain, which are based on human visual system characteristics. Under this theme, the following work has been done:(1) The background, the practical value and the research status of digital watermark technology have been introduced; The working principle of a watermarking system has been introduced. The basic thought, advantages and disadvantages of all kinds of typical watermarking embedding algorithm have been analyzed; Common attacks which are often used to against a watermark system have been introduced, the methods and various parameters of evaluating the watermarking system have been introduced.(2) This paper proposed a watermark preprocessing algorithm, which is based on Arnold transformation and chaotic encryption. The first step of the algorithm is to scramble the watermark image by Arnold transformation. Then the gained watermark will be encrypted by a chaotic sequence, to get the suitable watermark image for embedding. Before the preprocessing algorithm is proposed, the watermark image should be scrambled by Arnold transformation or preprocessed by the chaotic encryption. Then the single-method preprocessed watermark image should be embedded into a carrier image, change the number of Arnold scrambling or the initial value of chaotic sequences, acquire the PSNR curve which indicates the relationship of the two initial quantity and marked carrier image, get the best initial amount. The experiments proved that the ability against attack of watermark preprocessing algorithm of this paper is stronger than the ability of a single scrambling preprocessing algorithm.(3)Considering the characteristics of human visual system, the brightness and stripes of a carrier image, this paper proposed three adaptive watermark embedding algorithms, which are based on the blocked DCT transformation. Firstly, the carrier image should be divided into non-overlapping small pieces(size 8 * 8). Then the pieces should be divided into categories, based on the average brightness or the number of the stripe points of the small pieces or both of them. The watermark image data will be embedded into dc coefficients of DCT domain pieces with different intensity factor. The obtained small pieces are transformed by IDCT. Finally all the small pieces are put together to get the watermarked image. Proved by the experiment, the ability against attacks of the embedding algorithms proposed in this paper is much stronger than the existing embedded algorithm in DCT domain.(4) By attacking the image marked by the watermark, which is obtained by different embedded algorithms, the anamorphic watermark images which come from the attacked carrier image are extracted. The performance of different preprocessing and embedding algorithms are compared by the similarity of the two images.