An Image Watermarking Algorithm Based On HVS And Content Features

In the era of information-based, digital technology has brought people tremendous and unprecedented changes, through digital technology, efficiency and accuracy of information exchange have been greatly enhanced. However, everything has two sides, digital technology has not only taken a lot of convenience to our work and lives, but also brought us the information security issues, such as digital products copied for any number of times, transmitted, modified, or even maliciously modified and forged, this gives people a negative impact on life and huge losses on economy. Therefore, in the information era, the issues which protect the copyright information and ensure security information, has become the hot topic of scientific research. In this environment, digital watermarking technology, which can protect the copyright information, to ensure information security, has been created out. It is one of the important technologies which can protect digital products.Most existing watermarking algorithm based on image content has solved the contradiction between invisibility and robustness to some extent, but there are still some shortcomings, such as that the threshold of the image classification is difficult to determine, the effect of the image classification is not ideal and the location of the feature region is not precise enough, and so on. After analyzing the HVS and the features of image contents, this paper proposed an image watermarking algorithm which based on HVS and features of image contents, the algorithm uses standard deviation and coefficient ratio of low frequency energy to determine the classification, which makes it relatively easy to determine the threshold, then the algorithm uses a recursive decomposition of the quadtree to make the image classification and the location of feature region quite ideal, and make the efficiency of the algorithm have a certain promotion. By combining the texture and the brightness characteristics, this paper has also made certain achievements in resisting geometric attacks and JEPG compression.In this paper, we had deeply researched on the theoretical basis of digital watermarking technology. Comparing the mathematical formulas with the digital watermarking system, we can easily understand the basic model of digital watermarking technology; And then, we summarize the classification of the digital watermarking, the corresponding features and the current evaluation criteria of the watermark performance as well, and study the methods of the watermark attack, because the watermark and the watermark attack is a pair of contradiction, so the research on the watermark attack can enhance the robustness of the watermark.Through the depth research on characteristics of HVS and content features of the image, this paper proposes a robust image watermarking algorithm based on HVS and content features. In the part of watermark encryption, we presented a method of block scrambling transformation based on Arnold, the watermark image is scrambled on block, then scrambled on the overall image, the method uses a double encryption on the image watermark, and overcomes single-cycle scrambling of Arnold and greatly increased the difficulty of the watermark attack. In the part of selecting feature region, because the HVS has different sensitivity to image different regions, texture regions, which are generally more complex, can accommodate more watermark information, and information is hidden well, while watermark is embedded in the DC coefficient which has a stronger ability to resist attack, and has good robustness. The advantages of both can be well overcame the conflict between the watermark robustness and invisibility. First, using the average variance and Low-frequency energy ratio, image is classified as smooth, border and texture regions, then combined with image texture and brightness properties select a more reasonable region, we select the feature region with two factors of texture and brightness, so the method avoids the less ideal situation which is created by single characteristics factor. In the part of watermark embedding, using discrete cosine transform (DCT), we embed watermark into the DC coefficient, because the DC coefficient is few times more than the largest of AC coefficients, even the big times. In these large coefficients, embedded watermark signal against various attacks; After the image is subjected to various attacks, as long as the general outline of the image, which contains the watermark, exists, the algorithm can extract the watermark well, so the method has good robustness. Because watermark is embedded in the texture regions, and it is difficult to distinguish watermarked image with the original image by naked eye, the algorithm also has good invisibility.At last, using simulation tools to complete the algorithm, and simulate attack in the various situations, such as geometric attack, noise attack, JEPG compression and filtering attack. Using analysis of test results, the algorithm can extract meaningful information from the image which is subjected to various attacks, so it verifies the feasibility and correctness of the algorithm.