Performance Study of Facial Expression Recognition Using Convolutional Neural Network

Abstract

Facial expression depicts human emotions. Recognition of facial expression is used in various fields, such as for a better understanding of the customer’s desires during a home design consultation and to find out the pain suffered by a patient during medical treatment. This research explores deep learning techniques based on Convolutional Neural Network (CNN) on facial expression recognition. The three pre-trained CNN models, namely VGG16, Resnet50, and Senet50, are retrained using different learning rate values and optimization functions. Trials on The Extended Cohn-Kanade Dataset (CK +) consisting of 7 expression classes, namely anger, neutral, disgust, fear, joy, sadness, and surprise, produce the best accuracy of 97% obtained by the VGG16 architecture with Adam’s optimization function and learning rate of 0.001.