Automatic blurring of specific faces in video

Abstract

With the introduction of the General Data Protection Regulation (GDPR) into European Union law, it became more important than ever before to properly handle personal data. This is an issue for media companies which distribute large amounts of media containing identifiable people, which thus may require the subjects' permission for distribution. In this Master's thesis, I propose a solution which supports and facilitates compliance with GDPR regarding the distribution of video containing identifiable subjects by automatically blurring a select group of people in the videos. The proposed solution is a pipeline for detecting, identifying and blurring select faces, where the video frames are processed like individual images to detect and recognize faces, and the interrelatedness of adjacent frames in continuous videos is exploited to both to improve their prediction quality and running time. Each part of the pipeline is interchangeable and may be replaced individually, and the deployment of the entire pipeline has been automated. Aspects related to video processing, facial detection and facial recognition were explored for this purpose, and various existing tools and solutions were utilized.