Quadratic difference expansion based Reversible Watermarking for relational database

Abstract

With the increase in usage of databases over the internet, it is becoming difficult to recognize genuine database owner. To protect ownership, digital watermarking has emerged as an effective solution. However, embedding watermark into database would result in loss of data quality.  Most state-of-the-art methods introduce distortion into the original data to a large extent.  In this work, reversible watermarking technique using quadratic difference expansion has been proposed. First, the numeric attributes with highest pearson correlation coefficient are selected to reduce distortion in the database.  The watermark is then generated by extracting bits from the selected attributes using quadratic difference expansion. Watermarked database is obtained after the generated watermark is embedded into the original database. To resolve ownership conflict, the genuine owner can easily extract watermark from watermarked database and recover original database. Indian Liver Patient dataset is used to conduct experiments. Results show that the proposed method is 100% robust against insertion attack. In case of 90% modification and 50% deletion atatck, it is observed that around 50% of the watermark can be recovered. Moreover, watermark embedding results in only 0.02% change in the mean of original database. It ensures that distortion caused due to watermarking leads to lower effect on data quality.