Watermarking: How secure is the DM-QIM embedding technique?

Evaluation of the security of data hiding methods has gained considerable interest recently. The security of quantisation index modulation (QIM) based watermark embedding methods has previously been investigated for different classes of attacks including known-message attack, known-original attack and watermark-only attack. In this paper, an extreme case of the watermark-only attack scenario is discussed wherein only a single instance of a set of watermarked content is available. We challenge the security of the dither modulation (DM)-QIM based embedding method by examining feature space embeddings for time series and images, using a distribution-free algorithm to discover the primary key embedding parameters.