Robust database watermaking technique over numerical data

Abstract

Proving ownership rights on outsourced databases is a vital issue. This research proposed an improved algorithm for relational database watermarking to minimize data variation as well as robust watermarking. For minimizing data variation two minimizing methods are proposed, the first method avoids embedding watermark bits in attributes with small values, the second method flips one of the unmarked bits in marked attribute in order to obtain the nearest number to the original attribute value. For embedding watermark bits, this research transforms the watermark image into a two dimensional matrix elements and embeds a gray (black and white) image in a database table. Furthermore in extracting process, a recovering phase is added. Recovering phase finds the missing extracted watermark elements and increases the robustness, so that the watermark can be detected even in a small subset of a watermarked database. In extracting process, majority voting method is employed to retrieve the correct watermark. For obtaining accurate results this study uses three various images and three separate database tables with different sizes. By applying these methods, variation of data after applying watermark is extremely less and also an upper Correction Factor for watermark can be obtained. Detecting the watermark neither requires the original database nor the watermark. The results have illustrated that the proposed method are convincing compared to other related methods.