Impact of Features Extraction Technique on Emotion Recognition Using Deep Learning Model

Abstract

A computer system would have a far harder time recognizing emotions from facial expressions than a human would. In a variety of settings, especially for human-computer interaction, the social signal processing sub-field of identifying emotions based on facial expressions is applied. Many studies have examined automatic emotion recognition, the majority of which make use of machine learning techniques. It remains a challenging issue in computer vision to recognize basic emotions including happiness, contempt, anger, fear, surprise, and sadness. Deep learning has received more attention recently as a possible option for a number of real-world problems, containing emotion recognition. In this paper, we proposed the usage of a 1-Diminsion Convolutional Neural Network (1D-CNN) to recognize some of the basic emotions and employed a different type of preprocessing and feature extraction ways to show how these methods impacted the performance of the proposed CNN model. The experiments on the Crowd-sourced Emotional Multimodal Actors Dataset (CREMA-D) revealed a high accuracy rate of 99.8%.