Facial expression recognition plays a vital role in enabling machines to understand human emotions, with applications in healthcare, education, surveillance, and human-computer interaction.
This project explores emotion detection using Convolutional Neural Networks (CNNs) on a dataset of 48×48 grayscale facial images, categorized into seven emotion classes:
- Angry
- Disgust
- Fear
- Happy
- Sad
- Surprise
- Neutral
Two architectures are fine-tuned and compared:
- DenseNet-121 → Leverages transfer learning for higher accuracy and generalization.
- AlexNet → Lightweight, suitable for resource-constrained or real-time applications.
- Preprocessing of 48×48 grayscale images
- Implementation of CNN-based models for classification
- Comparison of AlexNet vs DenseNet-121
- Training and evaluation with accuracy metrics
- Visualization of predictions and results
The dataset consists of 48×48 pixel grayscale images categorized into seven emotions.
(Commonly used dataset: FER-2013).
- Healthcare: Patient emotion monitoring
- Education: Student engagement analysis
- Human-Computer Interaction
- Security and Surveillance