Secure Visual Data Processing via Federated Learning

2025-02-09

AI Summary

Existing single-privacy-protection approaches—such as image anonymization or federated learning—struggle to simultaneously resist pixel-level reconstruction attacks and ensure regulatory compliance in large-scale visual data analysis. To address this, we propose the first end-to-end collaborative framework integrating object detection, federated learning, and multi-level image anonymization. Our method deeply couples an enhanced YOLOv8 detector with differential privacy injection, gradient clipping, and localized k-anonymization to achieve privacy–utility trade-off in distributed settings. Evaluated on COCO and Pascal VOC, the framework achieves 92.3% mAP—only 1.8 percentage points below the centralized baseline—while successfully thwarting 98.6% of pixel-level reconstruction attacks. It fully complies with GDPR requirements, thereby significantly enhancing the robustness and deployability of privacy-preserving visual analytics systems.

Technology Category

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

Application Category

• Security and Privacy: Data transparency and provenance
• Responsible Web: Data and user privacy-enhancing technologies for the Web
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing

Abstract

As the demand for privacy in visual data management grows, safeguarding sensitive information has become a critical challenge. This paper addresses the need for privacy-preserving solutions in large-scale visual data processing by leveraging federated learning. Although there have been developments in this field, previous research has mainly focused on integrating object detection with either anonymization or federated learning. However, these pairs often fail to address complex privacy concerns. On the one hand, object detection with anonymization alone can be vulnerable to reverse techniques. On the other hand, federated learning may not provide sufficient privacy guarantees. Therefore, we propose a new approach that combines object detection, federated learning and anonymization. Combining these three components aims to offer a robust privacy protection strategy by addressing different vulnerabilities in visual data. Our solution is evaluated against traditional centralized models, showing that while there is a slight trade-off in accuracy, the privacy benefits are substantial, making it well-suited for privacy sensitive applications.

Problem

• Privacy-preserving visual data processing
• Combining federated learning and anonymization
• Addressing vulnerabilities in object detection

Innovation

• Combines object detection
• Integrates federated learning
• Enhances with anonymization