Research Of Print-scan Resilient Chinese Text Image Watermarking Technology Based On Character Skeleton

Since entering the era of information technology, the Internet in people’s daily life is increasingly deep and popular, and it brings great convenience to people. The electronic document is the most popular between all the communication media, because it is easy to copy and edit. With the development of the digital age, even many important documents such as social security, financial file has been digitized and stored. However because of the easier to copy and edit it also lead to many security problems about abusing unauthorized document, which makes it urgent to protect the copyright of the digital text. Digital watermark technology is a valid way to discourage illicit reproduction of documents by embedding the copyright information into texts. Although it is a good protector for the digital texts, the mature technique of printing and scanning makes it invalid to resist the print-scan attack. Therefore, it is worth studying print-scan resilient watermarking methods for the text images.In this paper, on account of the studying of robust watermark and print-scan resilient text watermarking, we proposed two kinds of print-scan resilient image watermark algorithms on Chinese texts. The key in solving the print-scan attack is to find a invariant during print-scan process. The first method proposed to fetch the difference between sum of character spaces or heights in different groups as the invariance. The watermark is embedded by slightly modulating the spaces or heights of characters, and the modulating is realized by shifting the characters left/right or up/down. Another one is the improvement of the former which combining the skeleton algorithm with the former one. The robustness of watermark is carried out by using the robustness of character skeletons after print-scan process. The main work of this paper includes: text image tilt correction algorithm, robust text watermarking algorithm based on shifting characters, robust text watermarking algorithm based on character skeletons. The specific research result is as follows:(1) The text image tilt correction algorithm:For the text tilt caused by printing and scanning, this paper finds that the sum of the inter-line blank spacing has a big gap before and after image tilt. Namely the bigger the tilt angle, the smaller the sum of the inter-line blank spacing. Thus, on account of this phenomenon, we can compute the maximum value of the sum by rotating the image step by step; and the angle rotated is the tilt angle. Then we rotate the image with the tilt angle by using a rotating function. Experiments show that the algorithm can compute the tilt angle in the precision of0.0001, which lay the foundation for the later character segmentation and watermark extraction.(2) Robust text watermarking algorithm based on shifting characters:The algorithm based on character moving depends on precise character segmentation. So this paper firstly proposed the character segmentation algorithm basing on the connected domain. Then on the basis of character segmentation algorithm the watermark is embed by shifting characters left/right or up/down. The experiments show that this kind of method has better robustness to print-scan attack and the embedding capacity is increased. But the edge pixels of characters would be affected by print-scan operation, causing wrong segmentation of characters, which will directly affect the value of character spaces and heights. So this kind of algorithm can be improved.(3) Robust text watermarking based on the character skeletons:Focusing on the problem mentioned above, we devote ourselves to find a better invariance during the print-scan process. Although the character skeletons only have one-pixel width after skeleton extraction, it retains the geometric and topological characteristics of the characters. And the skeletons of characters are basically the same before and after print-scan operation. Namely the character skeletons have robustness to print-scan attack. So we use the skeleton of characters to represent the invariant during print-scan process. Thus the robustness of watermarking can be achieved by combining the robustness of skeleton and the watermarking algorithm. The experiments show that this kind of algorithm has better extraction accuracy than the former one.