Correlation detection of asymmetric watermark Jin S. Seo and Chang D. Yoo Korea Advanced Institute of Science and Technology, Department of EECS, 373-1 Kusong-dong, Yusong-gu, Daejeon 305-701, Korea pobi@eeinfo.kaist.ac.kr, cdyoo@ee.kaist.ac.kr Abstract. This paper proposes a novel method to detect Furon’s asym- metric watermark by using a correlation detector that is mathematically tractable and simple. The performance of the proposed method is tested under various conditions. The experimental results matched the theo- retical results well, showing that the correlation detector can indeed be used for the detection of asymmetric watermark. The proposed detector is applied to both single and multiple bit embedded watermark. Bit er- ror rate (BER), obtained from the experiment, was compared to the one obtained from the theory. 1 Introduction With the advent of Internet, there has been an explosive growth in the use of digital media. Since digital media is easily reproduced and manipulated, anyone is potentially capable of incurring considerable financial loss to the media pro- ducers and content providers. In this respect, digital watermarking is essential. Most of the existing watermarking methods use symmetric key, that is to say the same key or pattern is used in the embedding and detection. Thus the secrecy of the key is shared by the embedder and detector. In situations where the detector must be available to the public, the secrecy can be divulged by tampering the detector. Based on public-key crypotography, T. Furon addresses this problem with asymmetric watermarking. In his work [1], the presence of filtered watermark that is considered as an output of a filtered random process is detected using only the knowledge of the magnitude of frequency response of the filter. In this paper, a simple and mathematically tractable detection method is proposed for the detection of asymmetric watermark. It is based on the the- ory of detecting a known signal in noisy channel. The known signal is the power spectrum of the embedded watermark, and the noise is the estimation error. The estimation error of the power spectrum is assumed to be additive, uncorrelated and Gaussian noise. By using periodogram averaging in the power spectrum estimation, the assumptions are satisfied. The advantages of periodogram aver- aging over periodogram used in [1] are the reduction of the variance of the power specturm estimate and computational load in estimating the power spectrum. The optimum threshold of the correlator output is set using the Neyman-Pearson H.-Y. Shum, M. Liao, and S.-F. Chang (Eds.): PCM 2001, LNCS 2195, pp. 638–645, 2001. c  Springer-Verlag Berlin Heidelberg 2001