Synthetic Biometrics

Motivation & Goals

Recently, Generative AI methods (e.g., GANs or diffusion models) have achieved considerable attention from the deep learning research community due to their significant contributions in image generation tasks. These networks can generate very realistic images, which could help to train better biometric systems (e.g., by increasing the number of samples in minority population to tackle bias, or to augment the amount of attacking samples for attack detection). However, the person's identity is frequently lost in this generation process, or artefacts reduce their usability for biometric recognition. This problem affects both facial and iris images, in the near-infrared and the visual spectra. 


  • Study the state-of-the-art in synthetic generation of biometric images for the chosen characteristic (e.g., face, visible spectrum iris, NIR iris)
  • Develop and evaluate new generation methods
  • Benchmark the developed methods against the state-of-the-art


Marta Gomez-Barrero (