BeFaced: A Game for Crowdsourcing Facial Expressions

Facial expression analysis systems often employ machine learning algorithms that depend a lot on the quality of the face database they are trained on. To train the systems robustly, the databases need to be large and contain high variability in terms of facial features, pose and illumination, amongst other variables. To achieve these goals, we have developed BeFaced, an iPad game based on the tried-and-tested tile-matching gameplay mechanic that has made many of such puzzle games successful. We created an alternative version of the tile-matching gameplay mechanic to include facial expressions as player input and aim to use the gameplay appeal to obtain a large database of natural and varied facial expressions in the wild. Moreover, the design of our gameplay mechanic allows for the ability to "request" the player for any type of expressions depending on what we show on the tiles. At a more abstract level, BeFaced investigates a novel method of using popular game mechanics to enable better affective algorithms.

Chek Tien Tan
Games Studio, University of Technology, Sydney