Practical analysis of watermarking capacity

Digital watermarking embedding is a hot research field of image processing. Digital watermarking is only possible because our vision system is not perfect. A number of applications have emerged such as copyright notification, time stamp and automated monitoring. There have been ingenious works on watermarking capacity, which all assume the attack distortion are Gaussian additive distribution. We argue that these results are beautiful but not practical. It is hard to classify different attacking methods into a uniform format (especially as Gaussian distribution). In the other hand, the capacity of a given image´s information is an interesting and important topic. Instead of conventional analysis based on non-precise assumption of attack distortion, this paper concentrates on the problem of how much information that image can carry without much visibility distortion. Our work is based on wavelet transform and mean-squared error since we believe they are representative measure for most work. Comparing with the former work on the watermarking capacity, this paper pays more attention to the influence on capacity of widely used technique such as spread spectrum. In the last part of this paper, we discuss future trends of new watermarking algorithm and their influence on capacity.