When compared with analog data, digital data has the advantage to be much more resistant to noise during its transmission or storage. However, this advantage also allows the errorfree copying of information. At the beginning of the information age, where informations are to be sold to multiple customers, this turns out to be a severe drawback.

In order to cope with this problem varios methods of embedding imperceptible watermarks into digital still images have been proposed so that by revealing the existence of the watermark the proprietor of the image could prove his ownership.

We think that communications theory is a suitable basis for the analysis of watermarking schemes. In our contribution we will present a watermarking scheme that generalizes many of today's known watermarking methods, and we will describe it as communications system. Analysis starts from the channel. We distinguish between two channel models: A honest customer applying standard image processing methods on one hand and an offender trying to destroy the watermark on the other. Standard image processing methods include lossy image compression, in particular according to JPEG. This limits the achievable data rate for the watermark. Since JPEG is based on the DCT transform, we analyze the DCT coefficients to obtain limits on the achievable data rate. Further we propose coding schemes to efficiently use the available data rate at a high reliability of the watermark data.

This work was done jointly with Hideki Imai

Back to Workshop Schedule