Applying Grammar Inference To Identify Generalized Patterns of Facial Expressions of Pain


Ute Schmid, Michael Siebers, Dominik Seuß, Miriam Kunz, Stefan Lautenbacher ;
Proceedings of the Eleventh International Conference on Grammatical Inference, PMLR 21:183-188, 2012.


We present an application of grammar inference in the domain of facial expression analysis. Typically, only sets of AUs which occur in a given time frame are used for expression analysis, the sequence in which these AUs occur is ignored. We wanted to explore whether the strucural patterns of AU appearances contain diagnostically relevant information. We applied alignment-based learning (ABL) to data of facial expressions of pain collected in a psychological study. To evaluate the quality of the induced grammars we applied cross-validation to estimate the generalization error. We can show that the learned grammars have reasonably high coverages for unseen pain sequences. However, the number of rules of the learned grammars cannot be reduced substantially without loss of generalization.

Related Material