Reversible Fragile Database Watermarking Technology using Difference Expansion Based on SVR Prediction

In this paper, the proposed scheme detects database tampering by embedding the important characteristics of the original database. The association rule of frequent pattern tree (FP-tree) data mining is utilized to determine the relationship existing among the protected attributes and others in the database. Meanwhile, support vector regression (SVR) is applied to predict each protected attribute value. By applying difference expansion (DE) and the differences between the original and predicted values, the owner can embed the digital watermark in the protected database. If the protected database is distorted, the SVR function can still predict the protected values. Then, an examination of the difference between original protected and predicted values allows for the extraction of the watermark. Data which has been tampered with can be found by comparing the original watermark with the extracted one. In this paper, FP-tree mining method is used to reduce SVR training time. Moreover, if the database has not been attacked then the proposed method can recover the original attribute values. When we extract watermark from the protected database and if this database has been tampered with, the proposed method can use the extracted watermark to authenticate and locate the tampered tuples. Therefore, the proposed database watermarking method can effectively authenticate the database integrity and protect the database.