Real-Time Emotion Recognition: An Improved Hybrid Approach for Classification Performance

Emotions, and more in detail facial emotions, play a crucial role in human communication. While for humans the recognition of facial states and their changes is automatic and performed in real-time, for machines the modeling and the emulation of this natural process through computer vision-based approaches are still a challenge, since real-time and automation system requirements negatively affect the accuracy in emotion detection processes. In this work, we propose an approach which improves the classification performance of our previous computer vision-based algorithm for facial feature extraction and automatic emotion recognition. The proposed approach integrates the previous one adding six geometrical and two appearance-based features, still meeting the real-time requirement. As result, we obtain an improved processing pipeline classifier (classification accuracy incremented up to 6-7%) which allows the recognition of eight facial emotions (six basic Ekman's emotions plus Contemptuous and Neutral).