Inpainting Assisted Self-embedding Watermarking Algorithms With Variable-length Recovery Codes

With the development of information technology, the digital multimedia applications have brought many conveniences to people’s work and life. But digital multimedia information can be easily tempered, forged and deleted during dissemination and utilization, which will make people misunderstand the events and even bring economic losses with the fake information.So the security issues of multimedia have attracted more and more attentions to study on. As one of the ways to protect the multimedia, the digital watermarking techniques embed the image information into the image self, which can detect tampered regions, and become a hot issue.Multimedia applications include image, video, radio, text and so on. This paper will mainly focus on image self-embedding watermarking. Self-embedding recovery, which is a special branch of watermarking techniques, can not only locate the tampered rejoins, but also restore them. This paper first will introduce and analyze existing self-embedding developments, later will research on how to balance the relationships among watermark capacity, the quality of watermarked and restored image by studying on the methods of generating and embedding watermark, and the security of algorithm. The contents and innovations involved in this paper are as follows:(1) In the existing self-embedding algorithms with variable-length recovery codes, the type information of a block is vulnerable to the malicious disruption. If the type information of the block is tampered, the recovery codes of all blocks following it will be incorrectly extracted, and the quality of recovered image will be decreased. To solve the above problems, a corrective method is proposed based on (7,4)hamming code and synchronous data. Hamming code and synchronous data are used to improve the accuracy of obtained type and extracted recovery codes, respectively. The experimental results indicate that the proposed method improves the ability against malicious attacks, and has better performance of recovered images.(2) Generating variable-length recovery codes with decreased amounts assisted by inpainting. When generating recovery codes, some blocks selected based on a small-scale-removing principle are skipped. So that the amounts of recovery codes embedded into the image are decreased. In reconstructing, a fast inpainting method is used to recover the contents of tampered regions accompanied with the extracted recovery bits. The experimental results indicate that the algorithm proposed by this paper has less revcovery data to generate, and a better quality of reconstructed image is achieved.(3) Improving the ability to against collage attack.Most existing self-embedding schemes don’t have ability to against collage attack. This paper divides the authentication data of block into two parts:one part is embedded into the block self and another into other block based on a secret key. When detecting tampered blocks, for the block whose reconstructed authentication data is inconsistent with the data extracted from other block by the key, we will determine its authenticity with the help of true blocks having the embedding relationships with it.In summary, the work of this paper mainly focus on how to develop the performance of self-embedding watermarking algorithm.First we improve the embedding method of variable-length recovery information based on (7,4) Hamming code and synchronous data.Then we give a method of generating variable-length recovery codes assisted by inpainting,which can decrease the amounts of recovery codes embedded into the image, and help to achieve a better quality of watermarked image. At last, collage attacks can be detected by proposed method, and a better quality of recovered image is achieved.