A variety of digital watermarking applications have emerged recently that require the design of systems for embedding one signal (the "embedded signal" or "watermark") within another signal (the "host signal"). We develop a framework for analyzing achievable performance trade-offs of these systems among robustness, distortion, and embedding rate. We also describe a recently introduced class of embedding methods, quantizationindex modulation (QIM), in which an ensemble of quantizers is constructed and information is embedded by quantizing the host signal with a quantizer associated with the watermark. We introduce an implementation of such a method called spread-transform dither modulation where the embedded information modulates the dither signal of a dithered quantizer, which quantizes projections of the host signal onto a spreading vector. We show that QIM systems have considerable performance advantages over previously proposed spread-spectrum and low-...
Brian Chen, Gregory W. Wornell