Citation

Popy, Kamrun Nesa and Tusha, Shahida Islam and Debnath, Tulika and Ahmed, Farhana and Rabby, Md. Sorowar Mahabub and Shah, Mohammad Shahin and Abdul Aziz, Nor Hidayati (2026) A Robust Ensemble-Based Framework for Facial Emotion Recognition Using Transfer Learning and Test-Time Augmentation. In: 2026 IEEE 2nd International Conference on Quantum Photonics, Artificial Intelligence & Networking (QPAIN), 16-18 April 2026, Chittagong, Bangladesh.

Text 9.pdf - Published Version Restricted to Repository staff only Download (632kB)

Abstract

Facial Emotion Recognition (FER) aims to automatically identify human emotion from facial images. However, FER remains challenging due to low-resolution images, class imbalance, and similarities between different expressions. In this paper, we propose an improved FER framework that combines image preprocessing, transfer learning, test-time augmentation (TTA), and ensemble learning techniques. Three pretrained convolutional neural networks-MobileNetV3, ResNet50, and InceptionV3-were fine-tuned on the FER2013 dataset. Before training, the facial images were enhanced using contrast-limited adaptive histogram equalization (CLAHE) and normalization to improve feature clarity. During testing, multiple augmented versions of each image were evaluated to increase prediction stability. The final prediction was obtained by averaging the outputs of all three models. Experimental results show that the best individual model (MobileNetV3) achieves 86.61% accuracy, whereas the proposed ensemble framework further improves the performance to 87.03% on FER2013. These results demonstrate that combining lightweight deep learning models with preprocessing and augmentation strategies can enhance the recognition accuracy while maintaining practical efficiencies.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Facial Emotion Recognition (FER), deep learn ing, convolutional neural networks (CNNs), transfer learning, MobileNet, ResNet50, InceptionV3, ensemble learning, test-time augmentation (TTA), FER2013 dataset, image preprocessing, real time emotion detection
Subjects:	T Technology > T Technology (General)
Divisions:	Faculty of Engineering and Technology (FET)
Depositing User:	Ms Suzilawati Abu Samah
Date Deposited:	31 Jul 2026 06:15
Last Modified:	31 Jul 2026 06:15
URII:	http://shdl.mmu.edu.my/id/eprint/16409

Downloads

Edit (login required)