Ensemble model using hybridization of angles and distances for emotion recognition (HADER)

Abstract

Integration of emotions to develop an integrated emotionally intelligent system is a tricky task. Human face is a complex entity which displays a variety of facial expressions. In this research study, an Ensemble model using hybridization of angles and distances for emotion recognition (HADER) is proposed. Proposed HADER forms a feature set containing a total of 37 angles and distances calculated using fiducial points. Hybridized feature set is submitted to ensemble model to predict the intensity of various emotions. HADER is validated on CK+ and KDEF datasets. Evaluation shows the increased accuracy results of proposed ensemble to be 99.630 and 98.639 on CK+ and KDEF dataset respectively over the existing geometric feature extraction based techniques attaining a highest accuracy of 90% on CK+.

Subject Classification:

• 68T45
• Machine vision and scene understanding

Keywords:

• Emotion Recognition
• Face Detection
• Geometric Features
• Support Vector Machine
• Logistic Regression
• Multilayer Perceptron