基于SVR预测的可逆数据库水印技术

REVERSIBLE DATABASE WATERMARKING TECHNOLOGY BASED ON SVR PREDICTION

频繁模式树(FP-tree)的关联规则是利用数据挖掘算法来确定受保护的属性和其他数据库之间存在的关系,支持向量回归方法(SVR)用来预测每个受保护的属性值。提出通过嵌入原始数据库的重要特征来检测数据库是否被篡改过的实现方法。通过对比差值扩展(DE)的原始值和预测值之间的差异,数据库管理员可以在受保护的数据库中嵌入数字水印,如果受保护的数据库是扭曲的,在SVR功能下仍然是可以预测保护值的。FPtree挖掘方法用于降低SVR的训练时间,如果该数据库已被篡改,我们可以从受保护的数据库中提取水印,来验证并定位篡改元组,恢复原来的属性值,如果该数据库没有遭到攻击,我们也可以利用水印保护数据库。因此,提出的数据库的电子水印方式可以有效地验证数据库的完整性和保护数据库的安全性。

The association rule of FP-tree （ relationship between protected property and other frequent pattern tree ） uses data mining algorithm to determine the databases. SVR （Support vector regression） method is used to predict each protected attribute value. We propose a method to detect whether the database has been tampered with by embedding the important feature of the original database. By comparing the difference between original and predicted values of the DE （difference expansion ）, the database administrator can embed digital watermark in a protected database. If the protected database is distorted, it is still possible to predict the protection value under the SVR function. In this paper, FP-tree mining method is used to reduce the training time of SVR. If the database has been tampered with, we can extract the watermark from the protected database to verify and locate the tampered tuples and restore the original attribute value. If the database is not under attack, we can also use the protection of database watermarking. Therefore, the database can effectively verify the integrity of the database and protect the security of the database.