Automated facial expression recognition using exemplar hybrid deep feature generation technique

The perception and recognition of emotional expressions provide essential information about individuals' social behavior. Therefore, decoding emotional expressions is very important. Facial expression recognition (FER) is one of the most frequently studied topics. An accurate FER model has four prime phases. (i) Facial areas are segmented from the face images. (ii) An exemplar deep feature-based model is proposed. Two pretrained deep models (AlexNet and MobileNetV2) are utilized as feature generators. By merging both pretrained networks, a feature generation function is presented. (iii) The most valuable 1000 features are selected by neighborhood component analysis (NCA). (iv) These 1000 features are selected on a support vector machine (SVM). We have developed our model using five FER corpora: TFEID, JAFFE, KDEF, CK+, and Oulu-CASIA. Our developed model is able to yield an accuracy of 97.01, 98.59, 96.54, 100, and 100%, using TFEID, JAFFE, KDEF, CK+, and Oulu-CASIA, respectively. The results obtained in this study showed that the proposed exemplar deep feature extraction approach has obtained high success rates in the automatic FER method using various databases.