FaceReader: an online facial expression recognition system

M. den Uyl, H. van Kuilenburg and E. Lebert

VicarVision, Amsterdam, The Netherlands

Facial expressions can contain a great deal of information and the desire to automatically extract this information has been continuously increasing. Several applications for automatic facial expression recognition can be found in the field of human-computer interaction. In every day human-to-human interaction, information is exchanged in a highly multi-modal way in which speech only plays a modest role. An effective automatic expression recognition system could take human-computer interaction to the next level.
Automatic expression analysis can be of particular relevance for a number of expression monitoring applications where it would be undesirable or even infeasible to manually annotate the available data. For instance, the reaction of people in test-panels could be automatically monitored and forensic investigation could benefit from a method to automatically detect signs of extreme emotions, fear or aggression as an early warning system.
The classification accuracy for an automatic system is however largely limited by the image quality, lighting conditions and the orientation of the depicted face. These problems can be partially overcome by using a holistic model based approach called the Active Appearance Model. In this approach, a face model is created in which lighting and pose are compactly represented. This model can then be analyzed and classified on a number of categories.
The FaceReader system to be demonstrated can classify facial expressions from one of the 'basic' emotional categories joy, anger, sadness, surprise, fear and disgust real-time and with remarkable accuracy. A performance of up to 90% correct has been achieved using frontal face images, making it one of the best performing systems in the field. The system can easily be trained to classify other facial expression categories, as is illustrated by its ability to individually evaluate several Facial Action Units. Furthermore, the system can also be used to classify other properties, such as the gender or age of the depicted person. The demonstration will consist of a poster presentation together with an online demonstration of our FaceReader system.

Paper presented at Measuring Behavior 2005 , 5^th International Conference on Methods and Techniques in Behavioral Research, 30 August - 2 September 2005, Wageningen, The Netherlands.

© 2005 Noldus Information Technology bv