FaceRec: An interactive framework for face recognition in video archives

Lisena, Pasquale; Laaksonen, Jorma; Troncy, Raphaël
DataTV-2021, 2nd International Workshop on Data-driven Personalisation of Television, at the ACM International Conference on Interactive Media Experiences (IMX 2021), 21-23 June 2021, New-York, USA (Virtual Conference)

Annotating the visual presence of a known person in a video is a hard and costly task, in particular when applied to large video corpora. The web is a massive source of visual information that can be exploited for detecting celebrities. In this work, we introduce
FaceRec, an AI-based system for automatically detecting faces of known but also  unknown people in a video. The system relies on a combination of state-of-the-art algorithms (MTCNN and FaceNet), applied on images crawled from web search engines. A tracking system links consecutive detection in order to adjust and correct the label predictions using a confidence-based voting mechanism. Furthermore, we add a clustering algorithm for the unlabelled faces, thus increasing the number of people that can be recognized. We evaluate our system that obtained high precision on datasets of both historical and recent videos. We release the complete framework as open-source at https://git.io/facerec.

Type:
Conference
City:
New-York
Date:
2021-06-21
Department:
Data Science
Eurecom Ref:
6549
Copyright:
© ACM, 2021. This is the author's version of the work. It is posted here by permission of ACM for your personal use. Not for redistribution. The definitive version was published in DataTV-2021, 2nd International Workshop on Data-driven Personalisation of Television, at the ACM International Conference on Interactive Media Experiences (IMX 2021), 21-23 June 2021, New-York, USA (Virtual Conference)

PERMALINK : https://www.eurecom.fr/publication/6549