A Nonlinear Projection Based Human Emotion Recognition Approach Employing Face Point Data

In human-machine interaction (HMI), human emotion recognition plays a critical role. Facial expression could be used as a means to capture human emotions. A facial expression is a nonverbal communication that can take the shape of facial muscle postures, revealing an individual’s emotional conditions. In this work, a hybrid feature extraction methodology, comprising a particle swarm optimization (PSO) guided polynomial space projection and an uniform manifold approximation and projection (UMAP) technique, has been applied to the in-house Austro-Mongoloid sub-race based facial information dataset. Further, to evaluate the performance of the proposed method random forest classifier has been utilized to classify facial emotions like happiness, sadness, surprise, fear, and anger. After successful implementation of the proposed methodology a blind test accuracy of 98% has been achieved that outperforms the other competitive methods.