Emotion Recognition Based on Occluded Facial Expressions

Facial expressions provide important indications about human emotions. The development of an automatic emotion recognition method is a challenging task and has applications in several domains of knowledge, such as behavior prediction, pattern recognition, entertainment, interpersonal relations and human-computer interactions. An automatic approach to emotion recognition based on facial expressions robust to occlusions is proposed and evaluated in this work. Robust Principal Component Analysis is employed to reconstruct the occluded facial expressions. Facial expressions are extracted through different features (Gabor Filters, Local Binary Patterns and Histogram of Oriented Gradients), which are used to recognize the expressions by Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) classifiers. Experiments conducted on three public datasets demonstrate the effectiveness of the proposed methodology.