Digital Watermarking

Watermarks are supposed to be imperceptible. This raises two important questions. How do we measure a watermark's perceptibility? And how can we embed a watermark in a Work such that it cannot be perceived? In this chapter, we describe the use of perceptual models to answer these questions.
In many applications, design trade-offs prevent a watermark from being imperceptible in all conceivable Works, viewed or heard under all conditions by all observers. However, imperceptibility should not be viewed as a binary condition. A watermark may have a higher or lower level of perceptibility, meaning that there is a greater or lesser likelihood that a given observer will perceive it. Section 7.1 discusses some basic experimental methodologies for measuring this likelihood and introduces the idea of estimating perceptibility with automated perceptual models.
Any automated perceptual model must account for a variety of perceptual phenomena, the most important of which are briefly described in Section 7.2. Many models of the human auditory system (HAS) [13, 202] and the human visual system (HVS) [287, 66, 173, 274, 281] have been constructed. In Section 7.3, we describe simple models of human audition and vision.
Finally, Section 7.4 describes several ways a perceptual model can be incorporated into a watermarking system to control the perceptibility of watermarks. These range from using the model for simple adjustment of a global embedding strength to distorting the embedded mark so as to maximize detectability for a given perceptual impact. Example image watermarking techniques, using the HVS...