Authors:
João Vitor Andrade Estrela
and
Wladmir Cardoso Brandão
Affiliation:
Department of Computer Science, Pontifical Catholic University of Minas Gerais (PUC Minas), Belo Hozizonte, Brazil
Keyword(s):
Children Identification, Face Detection, Face Recognition, Image Alignment, Neural Network.
Abstract:
Social project sponsors demand transparency in the application of donated resources. A challenge for nongovernmental organizations that support children is to provide proof of children’s participation in social project activities for sponsors. Additionally, the proof of participation by roll call or paper reports is much less convincing than automatic attendance checking by image analysis. Despite recent advances in face recognition, there is still room for improvement when algorithms are fed with only one instance of a person’s face, since that person can significantly change over the years, especially children. Furthermore, face recognition algorithms still struggle in special cases, e.g., when there are many people in different poses and the photos are taken under variant lighting conditions. In this article we propose a neural network based approach that exploits face detection, face recognition and image alignment algorithms to identify children in activity group photos, i.e., i
mages with many people performing activities, often on the move. Experiments show that the proposed approach is fast and identifies children in activity group photos with more than 90% accuracy.
(More)