Signal Processing 88 (2008) 117–130 Locally optimum detection for Barni’s multiplicative watermarking in DWT domain Jinwei Wang a,b,Ã , Guangjie Liu b , Yuewei Dai b , Jinsheng Sun b , Zhiquan Wang b , Shiguo Lian c a 28th Research Institute, CETC, Nanjing 210007, PR China b School of Automation, Nanjing University of Science and Technology, Nanjing 210094, PR China c France Telecom Research and Development, Beijing Center, Beijing 100080, PR China Received 21 November 2006; received in revised form 13 June 2007; accepted 10 July 2007 Available online 21 July 2007 Abstract With the increasing demands of copyright protection, watermarking technology has being paid more and more attention. In the design of a watermarking algorithm, a good watermark detection scheme can improve the detection rate, which pushes an increasing number of researchers to work with optimum detectors. In this paper, we propose a locally optimum using a locally most powerful test with respect to Barni’s multiplicative water marking which is based on the human visual system (HVS). In the proposed detection scheme, the probability density function of DWT coefficients is modeled using the generalized Gaussian distribution, and the decision threshold is obtained by the Neyman–Pearson (NP) criterion. Additionally, we prove that the existing correlation detection is a special case of the proposed locally optimum detection, and we provide an improved correlation threshold under the condition of locally optimum detection. Crown Copyright r 2007 Published by Elsevier B.V. All rights reserved. Keywords: Barni’s multiplicative watermarking; Locally optimum detection; HVS 1. Introduction Due to the wide application of digital multimedia, copyright protection becomes more and more urgent, which makes watermarking technology attract significant attention. Since good detection can further improve the successful detection rate of watermark, most of the publications are based on either the additive embedding rule or the multi- plicative one. The existing detectors can be classified into two types, i.e. blind detectors [1,2] and non- blind detectors [3–6]. However, blind detectors are more focused because of the large amount of digital multimedia. Let y i denote the watermarked transform coeffi- cient corresponding to the original coefficient x i , a i and w i denote the watermark strength and the watermark, respectively. For the additive rule y i ¼ x i þ a i  w i , (1) there are two types of blind detection methods. One is the optimum detection [3–6] and the locally optimum detection [3,7]. The other is correlation detection proposed by Barni et al. [8], which is ARTICLE IN PRESS www.elsevier.com/locate/sigpro 0165-1684/$ - see front matter Crown Copyright r 2007 Published by Elsevier B.V. All rights reserved. doi:10.1016/j.sigpro.2007.07.012 Ã Corresponding author. 28th Research Institute, CETC, Nanjing 210007, PR China. E-mail address: wjwei_2004@163.com (J. Wang).