Emotion Recognition Using Deep Neural Network with Vectorized Facial Features

Abstract

Emotion reveals valuable information regarding human communications. It is common to use facial expressions to express emotions during a conversation. Moreover, some interpersonal communication can be achieved using facial expressions only. Some facial expressions are universal, they express the same emotion across cultures. If a machine were able to interpret its user's facial expression correctly, it might be able to help its user more efficiently. In this paper, a novel vectorized facial feature for facial expression will be introduced. The vectorized facial feature can be used to build an DNN (Deep Neural Network) for emotion recognition. Using the proposed vectorized facial feature, the DNN can predict emotions with 84.33% accuracy. Nevertheless, compared with CNNs (Convolutional Neural Network) with similar performance, training such DNN requires less time and data.