BeFaced: A game for crowdsourcing facial expressions

Machine learning algorithms for facial expression analysis systems often depend on having a set of high quality face images as training examples. To train the systems robustly, the database needs to be large and images need to have high variability in terms of facial features, pose and illumination, amongst other variables. Unfortunately, collecting such databases is costly and time consuming. Moreover the current popular databases are mainly collected in artificial lab environments with relatively small population sizes. Crowdsourcing methods can alleviate some of these issues and are just starting to emerge in this area [McDuff et al. 2011]. However, current efforts mainly focus on tasks that require conscious effort, and only collect limited expression types.

Author: 
Chek Tien Tan
Daniel Rosser
Natalie Harrold
Presented At: 
SIGGRAPH Asia 2013 Symposium on Mobile Graphics and Interactive Applications, p. 97. ACM
Year: 
2013
Type: 
Conference Proceedings