The CarMen consortium is pleased to share another important scientific achievement from project partner Hochschule Darmstadt (H_DA).
Researchers from h_da presented their latest research entitled “Semi-Synthetic Iris Image Generation with Identity Preservation Using Latent Diffusion Models” at the 20th IEEE International Conference on Automatic Face and Gesture Recognition (FG2026), held in Kyoto, Japan.
The presented work addresses one of the key challenges in biometric research: generating synthetic Near-Infrared (NIR) iris images that preserve the biometric identity of the subject while supporting the development of reliable iris recognition systems.
Existing iris datasets often contain only a limited number of images per subject, while many synthetic image generation approaches fail to preserve identity information. To overcome these limitations, the proposed method introduces a novel latent diffusion model operating directly on the latent representation produced by an iris autoencoder. This enables the generation of semi-synthetic iris images while maintaining the identity information required for biometric recognition.
This research contributes directly to the objectives of the CarMen Horizon Europe project, which aims to develop trustworthy, privacy-preserving and secure biometric technologies for future border management and identity verification systems.
Presenting the research at IEEE FG2026 provided an excellent opportunity to exchange knowledge with the international biometrics community and further strengthen scientific collaboration within the field of trustworthy artificial intelligence.
Congratulations to the research team from Hochschule Darmstadt and all collaborators on this valuable scientific contribution!