Automated Human Facial Emotion Recognition System Using Depthwise Separable Convolutional Neural Network

Abstract

The automatic human facial emotion recognition (AHFER) system has its wide significant contribution in several disciplines, such as human-computer collaboration, human-robot interaction, and so on. Multiple research projects have been conducted regarding this topic because it is a challenging and interesting task, especially in the area of computer vision. The purpose of the work is to recognize facial emotions using a depthwise separable convolutional neural network (DS-CNN). Apart from that, a facial emotion dataset has been proposed, and splitting functions, intensity normalization, image cropping, and grayscale conversion have been used in data pre-processing. The AHFER system is capable of recognizing four types of emotions: happy, sad, angry, and neutral. The results of the experiment showed that the AHFER method is 99 percent accurate when training and 93 percent accurate when validating. Additionally, we determined the confusion matrix with precision, recall, and fl-score. A comparison between the DS-CNN and DNN models was performed. The DS-CNN model performed significantly better than the DNN model. The DS-CNN model could be improved in the future by including more facial emotion categories.