Digital Watermark Algorithm Improvement Research

As a main method of digital image copyright protection and a newinformation hiding method, digital image watermarking has been rapidly developed inrecent years. Robust watermark is a rather difficult research field. It is well known thatwatermark is robust enough if it is embedded in low frequency of transform domain. Inface of many transform domains, we are often puzzled where to embed our watermark.This article studies the main goal is: (SVM) tests with the recent data miningtechnology - support vector machine examines suitably inserts the position of thewatermark, and proposes the corresponding watermark algorithm. We study morewatermarks insert algorithm: The multilayered sensation training method as well as thisarticle proposed the support vector machine training method, uses similar inserts themethod to insert the watermark, through the robust test comparison, will obtain the quitegood test result later the continuation research to lay the foundation for us.In the first chapter of this paper, we give an overview of digital watermarking, including: development, classification, algorithms, principle, usual attack ways, detection and research topics.In the second chapter, we have elaborated some basic concepts of the supportvector machine, through two kind of basic classified: linear and non-linear supportvector machine to explain its basic principle. Should serve as the very good theoryfoundation for this article behind.In the third chapter, we through view of the support vector machine's owncharacteristic and the merit in the data class, the numeral watermark's owncharacteristic and the function. From above all, elaborated the support vector machinein the insert, in the extraction examination and in the attack examination processapplication in the watermark. Proposed based on the support vector machine numeralwatermark algorithm. In the fourth chapter, we propose a new robust and blind watermark algorithmbased on support vector machine. The algorithm realization principal part is: Carries onthe pseudo-random arrangement to the selection picture, selects the suitable sampletrain set and train set divides into the sample two steps of minute: one part uses for totrain the SVM, another part inserts the watermark, the length of the sample sequencedecided by the watermark's sequence. The experimental result showed this algorithmhas the extremely ideal robustness and the transparency. This kind watermark algorithmhas the practical value very much.Last part is summarize this article and bring forward the ameliorate way of thewatermark in the future.