The Watermark project addresses the growing problem of deepfakes and other forms of synthetic visual media by proposing an active, prevention-oriented solution through the development of a secure, encryption-protected, and robust image watermarking system. Instead of attempting to detect manipulations after dissemination, the project focuses on embedding imperceptible watermarks directly into images at the time of creation or publication. These watermarks serve to verify image authenticity and establish ownership, while preserving visual quality and remaining resilient to common, non-malicious image transformations.

The research and development activities were structured around three major tasks. The first task focuses on the integration of message encryption mechanisms into robust watermarking algorithms, aiming to combine asymmetric cryptography, error-correcting codes, and selective image region modification. The second task is dedicated to the development of a novel encryption-protected watermarking algorithm based on machine learning and computer vision techniques, particularly through the embedding and detection of watermarks in the latent space of images using Latent Diffusion Models. The third task culminates in the development of a web-based prototype platform offering image protection services for registered entities, based on token-derived watermarks, as well as public verification services to assess image authenticity and provenance.

---

---

---

---