Facial Emotion Recognition by Combining Deep Learning and Averaged Weight of Face-Regions

Expression of the face is one of the most natural features and most impressive signals for humans to convey their state of emotions and intentions. This paper proposes a new face emotion recognition method using a deep learning model along with a weighted average of face-regions. Firstly, the deep learning method is used to identify face image and facial landmarks. Then, the face image is divided into 7 facial regions (2 eyes, nose, forehead, mouth and 2 cheeks). Subsequently, Convolutional Neural Network (CNN) is trained to indicate each emotion on each sub-region of the entire face and a weighting assessment set for each set of identifications is used to combine with the predicted result. The experimental results on the FER2013 dataset has demonstrated that proposed method achieves an accuracy of 74.14 which outperforms the state-of-the-art methods.