Multimedia security subsumes research fields dealing with the enforcement of protection goals, in particular confidentiality and integrity, for and with the help of digital signals that represent parts of reality.
The wide acceptance of digital photography, audio and video technology coupled with ever better media processing software enables even ordinary users to create strikingly real-looking manipulations of digital images or audio recordings. However, it is utmost important that media data is reliable and authentic if they are used as basis for decision making, for example in court (photographic evidence), in science (photographs as empirical proofs) and also in the formation of public opinion (press photographs). As a result, the field of digital multimedia forensics has matured as scientific discipline. Forensic methods in general deal with the search and analysis of traces with the aim of reconstructing (criminal) activities as well as identifying culprits. Multimedia forensic methods, in particular, employ models of media content or digitalisation technology to detect significant deviations from the normal case. Serving as forensic indications, such deviations contain information that helps to verify the authenticity of media data. For example, digital images can be assigned to the digital camera or scanner used for their acquisition by means of statistical analysis. Related methods can discover traces of post-processing operations.
Our research at TU Dresden is focussed on improvements to image forensics. We also take a critical approach with regard to the limitations of such methods. In that we ask the question, how reliable the results of multimedia forensics are, even if the presumed counterfeiter is aware of the forensic tools and operates with due caution.
Steganography and Steganalysis
The literal meaning of steganography is "covered writing". Known since ancient times, the art deals with concealing secret messages among other inconspicuous data. Hiding the existence of a message adds an additional layer of secrecy on top of what can be achieved with cryptography. Suitable cover data for secret messages includes digital media, as they typically contain some random information (e.g. noise in images), in which small modifications are unnoticeable. The objective of steganalysis is to detect the use of steganography. In practice, a steganographic system is considered as secure unless an algorithm is found that can distinguish between steganographic objects and clean covers with higher probability of success than random guessing.
Our research on steganography and steganalysis is mutually complementary: Our insights from the construction of targeted steganalysis methods get incorporated in the design of new steganographic algorithms with improved security properties. The focus of our work is on image steganography (both JPEG and uncompressed) as well as exploiting additional information of the image acquisition process (scanner, digital camera). The probably best-known embedding algorithm developed in Dresden is called "F5".
Similar to steganography, digital watermarks are embedded into cover media imperceptibly. However, the objective is not to hide the existence of a message. Rather, robust digital watermarks are entangled with a host signal in such a way that they cannot be removed without damaging the medium severely. This requirement has consequences on the construction of suitable embedding methods. Digital watermarks can be applied, for instance, to embed authorship and copyright information.
Our research is concentrated on "breaking" digital watermarks; that is to remove them from given media. We are proud to have successfully participated in contests, which developers of watermarking technologies organise as test-beds for their new methods.