Mathematical Foundations of Computer Science and Computer Security

Comparison of correlation measures used for digital watermark detection

V. A. Anzhin, V. N. Trenkaev


Abstract: The paper addresses the problem of comparing correlation estimation methods used for correlation-based digital watermark detection. An algorithm is proposed for evaluating correlation estimation methods. The proposed algorithm is based on an experiment that counts the number of successful watermark detections and false positives in an image dataset after distortions introduced by filtering. Detection thresholds for different correlation estimation methods are determined so that even a slight decrease results in at least one false positive detection. The proposed algorithm was tested using linear correlation, normalized correlation, and correlation coefficient within a watermark embedding and detection framework based on the discrete cosine transform. The results of the experimental evaluation demonstrate that the normalized correlation achieves the highest detection performance.

Keywords: digital watermark, linear correlation, normalized correlation, correlation coefficient, content protection.

UDC: 004.056.55

DOI: 10.17223/2226308X/18/40