Real Time Emotion Classification Based on Convolution Neural Network and Facial Feature

Abstract

Health problems due to emotion disorder should not be taken lightly because they have worse effects on health. Emotion disorder leads to prolonged stress and causes mental fatigue. Therefore, emotions need to be classified as early as possible. This classification result can be utilized to determine a person's emotion and treatments required. In this paper, we proposed emotion classifier based on facial features. Here, we used Convolution Neural Network (CNN) to extract facial features from input images and classify them into 7 basic emotions: angry, sad, happy, neutral, fear, disgust, and surprise. Dept-wise separable convolution is applied, instead of the ordinary convolution in CNN, to reduce the number of trainable parameters so that the overall architecture of CNN can be made as simple as possible without compromising the accuracy. The simple architecture of the CNN allows us to make it work in real time. Our proposed method achieves an accuracy of 66% on 3.589 input images of FER2013 data set.