LFU:online

The authenticity of digital images, audio, video, and text is critical to a society that makes important decisions based on them. With the advent of generative AI, there is a real risk that malicious actors can lend credibility to misinformation by creating synthetic content that cannot be distinguished from real content. Digital watermarking has been proposed as an active authentication technique that can be used to attribute sources and trace information flows. The basic principles have been known and applied for decades, primarily to protect media content. More recent developments have extended this to generated media and trained models.

By the end of the course, students understand basic methods for watermarking and source attribution, and can apply them in the context of digital images and machine-learning models.

The lectures cover the following topics:

- Security notions for digital media
- Image acquisition and processing
- Generation of synthetic images ("deep fakes")
- Spread-spectrum watermarking
- Robustness, synchronization
- Source attribution with PRNU
- Deep-fake detection
- Watermarking in relation to neural networks
- Asymmetric watermarking
- Fingerprinting codes ("traitor tracing")
- Collusion atacks, anti-collusion methods

In the second half of the semester, students will be assigned a small programming project implementing concepts from digital watermarking.
At the end, students will present their solution in the class.

The lecture conveys theoretical knowledge, which is then used in oral and written tasks.
The exercises involve programming tasks as well as conceptual work and on-paper calculation.
The project connects the gained knowledge with practice, and further deepens the understanding.
The lecture and the literature are in English.

Continuous assessment based on participation, exercise sheets, written tests, and a submitted project.

Cox et al. Digital Watermarking and Steganography. Morgan Kaufmann, 2008.

Katzenbeisser, Petitcolas: Information Hiding. Artech House, 2016. Ch. 5-8.

Zhao et al.: SoK: Watermarking for AI-Generated Content. IEEE Symposium on Security and Privacy (S&P), San Francisco, May 2025: 2621-2639.

Prior knowledge in signal processing and cryptography is recommended.

Allocation of places in courses with a limited number of participants (PS, SE, VU, PJ)

In courses with a limited number of participants, course places are allocated as follows:

1. Students for whom the study duration would be extended due to the postponement are to be given priority.

2. If the criteria in no. 1 do not suffice, first, students for whom this course is part of a compulsory module are to be given priority, and second, students for whom this course is part of an elective module.

3. If the criteria in no. 1 and 2 do not suffice, the available places are drawn by random.

Curriculum MA Computer Science 2021W