Facial Expression Recognition Using Deep CNNs: A FER2013 Study | IJCT Volume 13 – Issue 3 | IJCT-V13I3P92
International Journal of Computer Techniques
ISSN 2394-2231
Author
Nouhaila Korchi
Abstract
Automatic recognition of facial expressions is a core challenge in affective computing and human-computer interaction. This paper presents a deep Convolutional Neural Network (DCNN) trained on a three-class sub- set of the FER2013 dataset, targeting the emotions of happiness, sadness, and neutral — the three ma- jority classes in the dataset. The proposed architec- ture consists of six convolutional layers organized in three blocks with increasing filter depth (64, 128, 256), batch normalization, ELU activations, max-pooling, and progressive dropout regularization. Online data augmentation and adaptive learning rate scheduling via ReduceLROnPlateau are employed to improve generalization. The model achieves an overall vali- dation accuracy of 82%, with per-class F1-scores of
0.92 (happiness), 0.74 (sadness), and 0.75 (neutral), evaluated on 2,127 validation samples. Analysis of the confusion matrix reveals that the model performs best on happiness, while sadness and neutral remain more challenging due to their visual similarity. These results demonstrate the effectiveness of deep CNN architectures for emotion-focused facial expression recognition.
Keywords
facial expression recognition, deep learning, convolutional neural networks, FER2013, affective computing, ELU activation, batch normal- ization.
Conclusion
This paper presented a deep CNN for three-class fa- cial expression recognition — happiness, sadness, and neutral — trained on a subset of the FER2013 dataset. The proposed DCNN, featuring ELU activations, pro- gressive dropout, batch normalization, and online data augmentation, achieved 82% overall validation accu- racy with a weighted F1-score of 0.82. The model performed best on happiness (F1=0.92) and showed expected difficulty distinguishing sadness from neutral (F1=0.74 and 0.75 respectively), consistent with the known visual similarity of these expressions.
These results confirm that well-regularized deep CNNs are effective for focused, class-balanced facial expression recognition tasks. The pipeline presented here — from preprocessing and augmentation to eval- uation and error analysis — provides a reproducible foundation for future extensions to broader emotion taxonomies and real-world deployment scenarios.
References
[1] P. Liu et al., “Deep learning for facial expression recognition: A step closer to real-life applica- tions,” IEEE Signal Processing Magazine, vol. 34, no. 6, pp. 27–35, 2017.
[2] C. R. Smith et al., “Comparative analysis of feature extraction techniques for facial expression recognition,” International Journal of Computer Vision, vol. 127, no. 6–7, pp. 703–724, 2019.
[3] X. Zhang et al., “Facial expression recognition based on deep convolutional neural networks,” Journal of Pattern Recognition Research, vol. 13, no. 2, pp. 95–105, 2018.
[4] I. J. Goodfellow et al., “Challenges in representa- tion learning: A report on three machine learning contests,” in Proc. International Conference on Neural Information Processing (ICONIP), pp. 117–124, 2013.
[5]M. F. Valstar et al., “Meta-analysis of the first facial expression recognition challenge,” IEEE Transactions on Systems, Man, and Cybernetics, Part B, vol. 42, no. 4, pp. 966–979, 2012.
[6]P. Ekman, “Constants across cultures in the face and emotion,” Journal of Personality and Social Psychology, vol. 17, no. 2, pp. 124–129, 1971.
[7]S. Khan et al., “Facial expression recognition using deep learning: A survey,” Computer Meth- ods and Programs in Biomedicine, vol. 153, pp. 93–113, 2017.
X. Liu and C.-L. Li, “Facial expression recogni- tion with convolutional neural networks: Coping with few data and the training sample order,” in Proc. IEEE CVPR Workshops, pp. 2024–2032, 201
How to Cite This Paper
Nouhaila Korchi (2026). Facial Expression Recognition Using Deep CNNs: A FER2013 Study. International Journal of Computer Techniques, 13(3). ISSN: 2394-2231.
Related Posts:
