Fairer Analysis and Demographically Balanced Face Generation for Fairer Face Verification

2024-12-04

IEEE Workshop/Winter Conference on Applications of Computer Vision

AI Summary

Face verification systems suffer from compounded demographic bias and privacy risks due to skewed training data. Method: This paper proposes a controllable and interpretable fair generative framework. Building upon DCFaces, it introduces the first end-to-end tunable synthetic face generation pipeline enabling demographically balanced sampling across gender and race. It further innovates a joint deep fairness statistical analysis method—integrating logit regression with ANOVA—to quantify bias sources and guide generative adjustments. Contribution/Results: Experiments demonstrate that, while preserving privacy, the framework reduces the equal opportunity (EO) gap across gender and race subgroups by 37%, improves overall verification accuracy by 0.8%, and significantly outperforms existing debiasing approaches.

Technology Category

• Computer Vision: Bias, Fairness & Privacy
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy
• Machine Learning: Ethics, Bias, and Fairness

Application Category

• Social Networks and Social Media: Fairness and bias in social network and social media analysis
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Security and Privacy: Data transparency and provenance

Abstract

Face recognition and verification are two computer vision tasks whose performances have advanced with the introduction of deep representations. However, ethical, legal, and technical challenges due to the sensitive nature of face data and biases in real-world training datasets hinder their development. Generative AI addresses privacy by creating fictitious identities, but fairness problems remain. Using the existing DCFace SOTA framework, we introduce a new controlled generation pipeline that improves fairness. Through classical fairness metrics and a proposed indepth statistical analysis based on logit models and ANOVA, we show that our generation pipeline improves fairness more than other bias mitigation approaches while slightly improving raw performance.

Problem

• Addressing fairness in face verification systems
• Mitigating biases in generative AI face datasets
• Improving demographic balance in synthetic face generation

Innovation

• Controlled generation pipeline for balanced faces
• Logit models and ANOVA for fairness analysis
• Improves fairness and slightly boosts performance