Synthetic Visual Biometrics (SynVisBio): From Generation to Defense

Recent breakthroughs in generative AI—including GANs, diffusion models, neural rendering, and multimodal foundation models—are reshaping the field of visual biometrics. These technologies enable the creation of highly realistic synthetic data for faces, iris, gait, and other modalities, offering scalable solutions to data scarcity, fairness, and privacy in biometric systems. At the same time, the growing fidelity and accessibility of synthetic content raise serious concerns about identity spoofing, deepfakes, and adversarial attacks on biometric recognition. As synthetic visual data becomes increasingly indistinguishable from real data, the need for robust defenses, ethical safeguards, and forensic tools is more urgent than ever.

Synthetic Visual Biometrics (SynVisBio): From Generation to Defense is a workshop organized at the international conference IEEE Big Data 2025. The workshop will be held as an online event. This workshop will bring together researchers and practitioners to explore the dual-use nature of synthetic visual biometrics—from generation and augmentation to detection and attribution—fostering responsible innovation at the intersection of generative AI, security, and biometric data science. 

Topics of interest include, but are not limited to:

• Generative models for biometrics: GANs, diffusion models, and neural rendering for faces, iris, gait, etc.; Identity-preserving vs. identity-agnostic generation; Prompt-driven visual synthesis and control using vision-language models 

• Detection of attacks on biometric systems: Detection of synthetic and manipulated biometric content; Robustness of detection systems to unseen generation methods; Evaluation benchmarks and generalization studies 

• Synthetic Data for Biometric Training: Use of synthetic data for improved recognition accuracy; Domain adaptation and domain generalization with synthetic inputs; Balancing real and synthetic data in supervised/unsupervised learning; Leveraging foundation models for data curation and augmentation 

• Security and Privacy of Synthetic Biometric Data: Privacy-preserving data generation; Information leakage and re-identification risks in synthetic datasets; Watermarking, provenance, and traceability of generated content 

• Fairness and Bias Mitigation: Synthetic data for reducing demographic bias in biometric systems; Fairness-aware generation and evaluation techniques; Auditing large-scale generative models for demographic and geographic bias 

• Forensic Analysis and Visual Attribution: Synthetic source attribution and fingerprinting; Explainable AI for synthetic content detection and auditing; 

• Dataset and Benchmark Development: New synthetic biometric datasets (face, gait, iris, etc.); Evaluation protocols and metrics for realism, identity control, and security 

Find details about paper submission HERE.