Facial Expression Recognition Application using Convolutional Neural Network

Abstract

Facial Expression Recognition (FER) is an important field of research in human-computer interaction, analysing facial expressions to understand human emotions. Traditional methods in FER suffer from low accuracy and rely on manual feature extraction, making them inefficient and less adaptable. To address these challenges, this study presents the development of a custom Convolutional Neural Network (CNN) model designed to classify seven facial expressions. CNNs are a deep learning technique widely used in image classification tasks due to their ability to efficiently extract relevant features from images. The research utilised the FER2013 dataset from Kaggle, which contains 35,887 labelled images, with 28,709 images for training and 7,178 images for testing. The custom CNN model achieved a maximum accuracy of 69% with a batch size of 32, a learning rate of 0.0001 and 400 training epochs. A confusion matrix analysis shows the model performs best with the “Happy” expression (88.3% true positive rate) and worst with the “Fear” expression (41.7% true positive rate). A tkinter-based application was created to allow users to leverage the CNN model to predict the facial expression of any given input image and display the breakdown of percentages for all seven predicted expressions.