User-Centric Video Privacy Preservation: Automated Face Detection, Recognition, and Blurring for PDPA and Beyond

Video anonymization has become essential for protecting personal data across surveillance, social media, education, and research domains. This paper presents a user-centered anonymization framework that combines frame skipping with clustering-based face selection to reduce computational cost while maintaining anonymization fidelity. Experimental results show that skipping up to five frames yields significant efficiency gains before performance gains plateau, while clustering effectively suppresses redundant detections and minimizes manual annotation. A usability evaluation with diverse participants confirms that the system delivers both acceptable anonymization quality and intuitive operator interaction, underscoring its practicality for real-world deployments. Findings also reveal a trade-off between speed and detection accuracy at higher skip intervals, motivating future research on adaptive thresholds. Beyond conventional blurring, the framework supports creative overlays such as sunglasses, stylized masks, or context-aware rendering, broadening its applicability to live streaming, secure teleconferencing, e-learning platforms, and PDPA/GDPR-compliant workflows. By aligning privacy protection, computational efficiency, and usability, this work establishes a foundation for scalable and compliance-aware video anonymization systems.