Feature Fusion Based Convolutional Neural Network for Facial Expression Recognition

Face expressions are straight and significant social signals in communication of people, with another gestures and convey significant nonverbal emotion of cues in relation of interpersonal. The existing techniques has limitations of less recognition accuracy and doesn't fuse features. To overcome these limitations, Feature Fusion based Convolutional Neural Network (FF-CNN) method is proposed to recognize facial expression. The dataset utilized for the research is RAF-DB dataset and proposed method utilizes image branch for extracting middle and huge-level features from entire input image. Patch branch utilized for extracting local features from 16 image patches in actual images. L2 norm-based feature selection is performed towards acquire much discriminatory local features. Proposed method is evaluated by performance metrics of accuracy, precision and recall. The proposed method attained high accuracy of 91.23% which is superior when compared with other existing methods like FF-Artificial Neural Network (FF-ANN), FF-K-Nearest Neighbour (FF-KNN) and FF-Recurrent Neural Network (FF-RNN).