Research On Relational Database Watermarking Based On Content

All kinds of tamper made to multimedia information in the process of transmission or usage result in the suspicion to the integrity of digital media(digital image, audio, video and database, etc.) and the authenticity of content. Because encrypt technology can not provide further protection for deciphered data, fragile watermarking technology make up this defect and can be used for authenticity verification of digital media. In this paper, an improved strategy and its algorithm are put forward for two defects in the fragile watermarking algorithm proposed by Guo, et al., and improved effects are verified by using emulated experiment.The algorithm mainly consists of two basic subroutines: watermark embedding and watermark extraction and detection. Firstly, the tuples of database is divided into different groups and sorted to change a disordered state between tuples into a relative ordered state. Based on this result, two kinds of watermarks are constructed for each attribute value in a group: attribute watermark and tuple watermark. Attribute watermark is generated according to embedding key and all values of the same attribute, while for tuple watermark, it is formed according to the same embedding key and all attribute values of the same tuple, and then the two kinds of watermarks are translated into binary code. To embed watermarks, the least significant bit of a relational data is set to one bit of attribute watermark, and the next least significant bit of this relational data is set to one bit of tuple watermark. Because attribute watermark and tuple watermark respectively represent the column and row of data. In this way, the embedded watermarks actually form a watermark matrix, which helps to detect tamper to database. In watermark extraction and detection, the original watermark are constructed by using the method in embedding watermark, and the watermark information made of two bits respectively extracted from the least significant bit and the next significant are used to construct attribute watermark and tuple watermark The next step is to compare the original watermark with extracted watermark. At last, in the experiments for tamper attack of database including tuple insertion, tuple deletion, attribute value modification and attribute deletion, some key problems of the algorithm are discussed. Emulated experiment results show that embedded watermark is fragile. In watermark detection, the algorithm may be used for blind extraction because the original database is not necessary. The proposed fragile watermarking scheme can localize and characterize three kinds of operations: tuple insertion, attribute value modification and tuple deletion. One advanced characteristic of the algorithm has good capability of tamper detection to the end attribute deletion of database. The analysis of allowable error implies that the allowable error of relational data of database is non-uniform. Watermark embedded into database must not affect the availability of data according to the application of relational data.