Synthetic face data similarity does not always mean privacy leaks
Not All Similarity Is Leakage: Distribution-Aware Identity Leakage Assessment in Synthetic Face Datasets
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
Synthetic facial images can help protect privacy, but sometimes they look too much like real people used to train the models, causing privacy issues. The authors found that previous ways of spotting this problem had flaws, so they created a better method called Scaled Overlap to measure true identity copying more accurately. Their findings show that some synthetic datasets have more identity leakage than others and that most synthetic images mix features from different people instead of copying just one. This helps improve how we trust and use synthetic face data.
What this means in practice
- •For facial recognition developers: Assess and reduce identity leakage risks when training facial recognition systems with synthetic datasets using the proposed Scaled Overlap metric.
- •For synthetic data providers: Evaluate and certify synthetic face datasets for privacy by measuring true identity overlap and leakage beyond simple similarity checks.
Authors
Paweł Borsukiewicz, Daniele Lunghi, Wendkûuni C. Ouédraogo, Fadi Boutros, Jacques Klein, Tegawendé F. Bissyandé
Abstract
Synthetic data provides the means for more ethical and less biased facial recognition. Nonetheless, the limitations of generative models and the conditioning process they rely on pose a risk of identity leakage, resulting in synthetic samples that are overly similar to those seen in a generator's training set, mitigating privacy benefits. In this work, we show that previously proposed approaches are affected by two key limitations: high classifier dependence and a tendency to identify leakage with respect to real datasets that were not used in training. To address these shortcomings, we propose a novel metric called Scaled Overlap (SO), which adapts a non-parametric data-copying test for facial recognition and extends it to the identity level. By leveraging a multi-model setup that compensates for errors in individual classifiers, it provides insights into the true strength of identity leakage. Our analysis identifies leaking datasets (CemiFace, SFace2, IDNet) with the most acute leakage introduced by generative model conditioning in CemiFace, demonstrates that real datasets do not contribute equally to generator training, and shows that the majority of synthetic images are drawn from multiple identities rather than relying on a single identity. Ultimately, our study provides new insights into the quality of synthetic datasets and gives more reliable techniques to help the community ensure the trustworthiness of synthetic data.