Maximum likelihood watermark detection in absolute domain using Weibull model

Abstract

Maximum Likelihood (ML) detection scheme is regarded as one of key components of many blind image watermarking algorithms in various transform domains. In ML detection, a proper Probability Distribution Function (PDF) such as the Generalized Gaussian Distribution (GGD) is usually required to model the statistical characteristics of the transform coefficients of the watermarked images. However in some cases, the GGD is not the most suitable model due to its limitation in modeling the pulse-shape distribution. In this paper, we propose a novel ML detection scheme. By performing ML detection in the absolute domain, we utilize the Weibull distribution, a special case of the Generalized Gamma distribution, to model the absolute transform coefficients. The experimental results demonstrate that the proposed detection scheme outperforms the conventional ones in both DWT and CT domain for natural images. Furthermore it improves the watermark detection rates averagely by 75.03% for Computer Graphic (CG) images compared with the conventional algorithm.