Human facial emotion recognition using improved black hole based extreme learning machine

Facial Emotion Recognition (FER) plays an essential role in human-to-human communication and human-to-machine interaction. Based on the analysis of the facial expressions, the machine can understand the emotional status of the human and take suitable actions. A huge amount of works was done by researchers for decades to build FER systems that are able to discriminate facial emotion features and identify their categories. In this paper, a novel FER framework is suggested to overcome the drawbacks of the previous systems. The Extreme Learning Machine (ELM) universal approximation characteristic along with the Improved Black Hole algorithm global search ability are combined and used to classify the facial images. The Linear Discriminant Analysis (LDA) and Principal Component Analysis (PCA) are utilized to reduce the dimensions of the face images and keep the most discriminative features before feeding them into our system. The proposed system is evaluated over Japanese female facial expression (JAFFE), Karolinska directed emotional faces (KDEF), and extended Cohn-Kanade datasets (CK+), and succeeded to achieve an accuracy of more than 90% over all the datasets. The experiments are extended by testing the proposed system over our own designed facial dataset where the acquired accuracy of the LDA-BH-ELM approach reached 77%, 80% over CK+, KDEF datasets respectively. The comparison of results with the previous methods proved the efficacy and effectiveness of the proposed system, and its ability to achieve outstanding performance.