Error-control watermarking

Digital watermarking enables us to hide a message in another data file, such as an image. For instance, an artist selling digital copies of his pictures on-line, can embed a watermark identifying himself as the copyright owner. Watermarks are also used for authentication and self-restoration of images. Take photos from a crime scene for instance. Someone may try to tamper with the evidence, say replacing the section showing the unlicensed gun with a cup of tea. The watermark, containing information about the original image, can demonstrate that a change has been made, or even recover the original image.

A main challenge is to make the watermark robust. Normal processes such as analogue transmission and image processing result in (imperceptible) noise to the image, and this can damage or destroy the watermark. In the criminal scenarios above, the criminal will obviously attempt to remove the watermark in order to get away with the crime. Known solutions exist resisting a wide range of attacks, but combinations of wide ranges of attacks remain a challenge. The clever criminal will obviously tailor his attack to the application, and is likely to find a weakness.

Our vision is a new watermarking scheme which is secure against all known attacks.

We propose research towards this long-term goal by:

1. Using know results from error-control coding to develop a novel watermarking system which is robust to so-called local geometric attacks and cropping
2. Defining a coding-theoretic model to facilitate further improvement of error-control codes for watermarking applications
3. Exploring ways to combine features from different known solutions to get robustness against more attacks.