Correlation coefficient analysis of original and watermarked dataset for data usability

Abstract

Recent literature has presented various watermarking techniques for copyright protection of digital assets. These watermarking techniques focus on algorithms along with proof of robustness against various malicious attacks but do not evaluate these techniques for effectiveness. Datasets, obtained after watermarking, from relational databases are available on public networks for sale and outsourcing to legitimate users. These watermarked datasets are assumed to be accessible by malicious users for nefarious objectives. It is essential to analyze correlation coefficients of original and watermarked datasets to determine the usability of a watermarked dataset, if it has been pirated by a malicious user. In this article, we have put forward a framework for correlation analysis of original and watermarked datasets to determine the usability of a pirated watermarked dataset. The strength of correlation between original and watermarked dataset is measured by statistical tests such as correlation coefficient, hypothesis test, t-test evaluation, and regression analysis. The watermarked dataset is obtained from row replication technique for watermarking relational databases. Experimental analysis has shown an increase in variation among original and watermarked datasets and decrease in correlation as the number of watermarks is increased. We conclude that low correlation indicates the strength of the watermarking technique and also decreases the usability of a watermarked dataset as it does not provide any useful information to the malicious user.