Holistic and Component-Based Approaches in Facial Expression Recognition for Performing Arts

Facial expression recognition (FER) is widely used in applications like neurosciences, affective computing, human-machine interaction, behavioral analysis, etc. However, an application where the use of FER remains completely unexplored is performing arts. It is observed that the classic person independent FER system does not produce satisfactory results on Indian Classical Dance specific dataset due to the different ways of representation of facial expressions in real life and in performing arts. This is our effort to introduce a novel application where classic FER systems (person independent) are used to classify the facial expressions used in Indian Classical Dance. For the person independent system, Local Binary Patterns (LBP) are used as facial feature descriptors and the expressions are classified using Linear SVM. Japanese Female Facial Expressions (JAFFE) is used as a standard dataset to test the FER system and a comparable recognition rate of 83.7% is achieved. The same system is then tested on our dance specific dataset. A rule-based person-dependent system is designed for classification with the help of geometric distances between facial components. The facial component-based approach achieves the recognition rate of 85% on our dataset with 7 classes. This work shows promising results and triggers the exploration of the use of facial expression analysis in performing arts.