Research On Digital Image Watermarking Techniques

With the rapid development of computer and Internet techniques, the multimedia security becomes an important and challenging research topic. In this situation digital watermarking techniques achieve wide attention and become one of the most rapidly-developing subjects. Research on key techniques in watermarking field has theoretical and practical significance.The introduction of the classification, applications, present research situation, performance analysis and the potential attack types is given first. Then focusing on digital image, the fundamental principle of watermarking, including generation of watermark signals, watermark embedding and watermark detection is systematically described. Some properties of human vision system and its application in watermark are introduced. Meanwhile, the detailed researches on common mathematical transforms of watermark, such as Fourier transform, cosine transform and especially wavelet transform are carried out.The present study focuses on invisibility and robustness resistant to geometric attacks of image watermark. The main research topics are as follows:1. Watermark algorithm based on image edge characteristicMaking use of image edge characteristic, watermark is embedded. To get close continuous edge with good location and measure so as to increase robustness of algorithm, and in view of the disadvantage of current algorithms of edge detection, image edge geometrical model is attained on the basis of minimal oscillation principle and zero normal curvature. This model is simplified and improved with the use of grads vector. The image is divided into 8×8 blocks before watermark is embedded. According to the numbers of image edge points, watermark is embedded on the DC of DCT coefficients. Making the best of human vision system, the algorithm is robust.2. Watermark algorithm based on image shape-specific pointsBecause geometrical invariance of image edge is influenced by some factors such as noise, shape-specific points with rigorous geometrical invariance, found in the image edge points, compose a referenced point set. In the tested image, we can get shape-specific point set with the same method. The geometric transform parameters can be solved by matching the two point set. To get blind detection, the ordinates of the referenced points are coded into binary bit string called secret key. Compared with the common watermark schemes based on image edge, our algorithm has excellent performance resistant to geometric attacks by using the good geometric characteristic.3. Watermark algorithm based on region with salient featureAccording to the human vision attention model, there are some regions of interest which include salient feature. To extract these regions accurately, image segmentation is first realized with level set method, and the segmentation results are classified in the wavelet domain. The watermark is embedded onto the important wavelet coefficients to ensure invisibility and robustness. Experimental results have shown the embedded watermark is perceptually invisible and robust to various attacks.4. Watermark algorithm based on Radon and Fourier-Mellin transformsIn view of the disadvantage of watermark scheme based on Fourier-Mellin transform, a watermark algorithm is proposed by combining Radon transform and analytical Fourier-Mellin transform to get the feature of geometrical invariance which is applied in the embedment of watermark. Because the feature of invariance has theoretical basis and avoids the disadvantage of Fourier-Mellin transform, therefore, the robustness of watermark algorithm is greatly increased.