Invisible Watermarks | Technology
There are various ways to embed digital watermarks in
image files. In this interview, Martin Steinebach from the
Fraunhofer Institute for Secure Information Technology
explains the pros and cons of the most common
watermarking methods and suggests some other ways to
protect digital image content.What other methods compete with digital watermarks for the
protection of media data?Martin Steinebach:There are various alternatives to digital
watermarks. If you want to prevent data from being copied, then
proprietary file formats or a Digital Rights Management (DRM)
system are great techniques to use – in other words, you either
keep your data format secret or you encrypt it. If you want to
find stolen data, robust hash techniques – also known as
fingerprinting – are an excellent alternative. In this case, a heavily
simplified description of the media (similar to a checksum) is used
as a reference. This description data can then be used to compare
found data with the original.What are the advantages and disadvantages of digital watermarks
compared with other techniques?MS:Watermarks are not designed to prevent copying, but rather
to help you find and retrieve data that has been stolen. They are
often used as a deterrent when it comes to enforcing copyright
law. If you want to prevent your data from being copied in the first
place, you will have to use DRM or some other copy protection
system. The main advantage of watermarks is that they are fully
transparent for the end user, and they do not alter the data format
or its quality. Unlike robust hashing, watermarks can mark files
individually – in other words, you can produce multiple versions of
a single image. Robust hashing can distinguish between different
image files, but not between individual copies of a single image.
however, it is often more robust than watermarking and is capable
of surviving stronger compression and scaling processes.Can you briefly describe the technical aspects of how a digital
watermark works?MS:There are virtually limitless ways to embed a digital
watermark. A common method involves what is called
patchworking in a frequency domain. Here, a Fourier transform is
used to transform an image from its original pixel-based space
into a ‘frequency domain’ in which the image exists in the form
of a set of frequencies that relate to the individual color channels.
A secret key is then used to create frequency pairs and alter the
strength of the frequencies so that they form a specific
mathematical relationship to one another, whereby the
relationship represents a single bit of the resulting watermark.
A ‘zero’ or a ‘one’ bit is then embedded in the image, depending
on which of the two frequencies is stronger. Once all these bits
have been embedded, the watermarked image is transformed
back to its original pixel space.How does this work in practical terms for a potential customer?MS:We usually license our methodology to the owners of online
shops who store their digital wares in a secure server environment
where watermarks are embedded. Only the shop owner has access
to the secret key that is used to protect them. The watermarked
file a customer receives is indistinguishable from an unmarked
one. The shop owner’s terms of business advise the customer of
the fact that the image can be traced. If a shop owner finds an
illegal copy of an image, the watermark can be read and the
original customer can be traced.How can everyday users find out if an image contains a Fraunhofer
watermark?MS:In short, they can’t! Identifying one of our watermarks requires
knowledge of the process and access to the customer’s individual
key, which should, of course, remain secret. This is one of the
major advantages of watermarks over cryptography – a potential
attacker can never be sure whether a watermark has been
successfully deleted because there is no sure way of checking. If
you have cracked cryptographically protected data, you know you
have succeeded because you can see or use the data in its original
form. If we licenses our methodology to a service provider who
then uses it to provide customers with a watermarking service, the
provider will have to provide a way of reading watermarked data.
We always license embedding and reading technology together,
although it has to be said that we only sell licenses to companies,
not to individuals.Where do you think the development of digital watermarks is
headed?MS:Watermarking methods are constantly under development,
and we are still a long way from discovering the perfect technique.
There a still many ideas from the world of graphics processing that
we can build into our processes to make them more robust. One
major challenge is speeding up our processes to keep pace with
the increasing size of the images in everyday use. Our aim is to
adapt methods already used in the movie and music industries for
use with digital images. One of our long-term projects is data
integrity protection. We are currently using our CASED security
cluster to investigate ways of using watermarks as a kind of secure
coating for digital images that makes changes to their content
visible. We have already come up with some great leads, but so far,
none of the resulting methods are secure enough for use in
real-world applications.Martin Steinebach has a PhD in informatics and is head of the
Department of Media Security and Data Forensics at the Fraunhofer
Institute for Secure Information Technology (SIT). Much of his work
involves developing digital watermarks for audio, video, image and
document data. The Fraunhofer Institute licenses its processes to
industrial and media customers, as well as to the public sector.Interview: How Do Digital Watermarks Work?