Supervisor
Kislay Raj
Programme
MSc in Data Analytics
Subject
Computer Science
Abstract
Facial expression recognition (FER) is a highly relevant problem in the field of computer vision, with applications in various areas. Although this topic has made significant progress in recent years with deep learning, facial expression recognition continues to face significant challenges when faced with a noisy, imbalanced database, and low-resolution images, as is the case with the widely used FER2013 database. This work investigates the usability of different approaches based on Convolutional Neural Networks (CNNs) for the task of facial expression classification.
First, a custom CNN was developed and trained from scratch, using FER2013 as database. Subsequently, three well-known pre-trained architectures (VGG16, ResNet50, and MobileNetV2) were fine-tuned using transfer learning techniques in two phases. Evaluations of the different approaches were conducted under the same experimental protocol, seeking standardization across the design, using metrics such as accuracy, precision, recall, F1-score, and confusion matrix analysis.
The results indicate that the custom CNN achieved competitive results, while the VGG16 model achieved the best results, demonstrating greater balance between classes. MobileNetV2 performed well considering the computational cost, and ResNet50 achieved intermediate but no less relevant results. The LIME technique was also used to analyse interpretability and the most relevant regions in the expression classification task.
The study demonstrates that a custom CNN can be viable and efficient in resource-limited settings, even if pre-trained models perform slightly better. A critical comparison of the approaches presented in this study highlights advantages, limitations, and opportunities for improvement, potentially contributing to the advancement of more robust and reliable solutions for facial expression recognition tasks.
Date of Award
2025
Full Publication Date
2025
Access Rights
open access
Document Type
Capstone Project
Resource Type
thesis
Recommended Citation
Soares da Costa, G.
(2025) A Comparative Analysis of the Performance of Customised and Pre-Trained Convolutional Neural Networks in the Classification of Facial Expressions. CCT College Dublin.
DOI: https://doi.org/10.63227/652.299.80