Interaction design of a semi-automatic video face annotation system

Abstract

In this paper, we propose an intelligent system that allows people to annotate face tracks of video content. To reduce the workload, we adopt two strategies in the system: 1) visually similar face tracks could be grouped together to reduce human reaction time; 2) face models could be learned to automatically recognize celebrity identities, leaving annotator only simple judgement tasks. With more precise face models, the recognized results require much less time for confirmation. Altogether, these strategies significantly reduce the workload of human annotation/confirmation. Experiments on a very large video repository prove the efficiency and effectiveness of the proposed system. © 2014 Springer International Publishing.