Low Distortion Database Watermarking Based on Attribute Prediction

In the digital era, databases are increasingly used in the fields of healthcare, education, economics, and trade, but the open sharing of high-value databases also brings the risk of copyright infringement. Database watermarking technology proves copyright ownership by embedding digital watermarks in databases. Existing database watermarking schemes mostly embed watermarks by modifying the attribute values, resulting in data distortion to a certain degree. In the existing methods, there is a category of database watermarking methods that use two separate carrier data for watermark embedding, which enhances the watermark capacity and invisibility. Through theoretical analysis, it is found that the data distortion caused by this category of method is proportional to the difference between the carrier data. Based on this conclusion, this paper proposes a low-distortion database watermarking scheme based on attribute prediction to reduce the data distortion caused by the watermark embedding process. For each selected attribute value y, we use the random forest model to obtain its predicted value <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\hat{y}$</tex> and use <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$y$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\hat{y}$</tex> as two carrier data for watermarking embedding, since the prediction error (the difference between the attribute value and its predicted value) is small, the distortion caused by watermark embedding process should also be small. Experimental results confirm our theoretical analysis and the distortion caused by the proposed scheme is only 30 % of the current optimal database watermarking scheme.