Facial Recognition and Emotion Classification

Date: 30-04-2024
Author: Rizwan

Project Description

This project builds a real-time emotion detection system using facial recognition and machine learning to identify various emotions from webcam feeds. It displays the detected emotions with corresponding images and names on the screen. Additionally, a dynamic score bar graph shows confidence levels for each detected emotion, enhancing user interaction and system understanding. The system is trained to modulate webcam images based on the subject's expressions in real time and pre-recorded videos using a variety of emotions including 'angry', 'disgust', 'fear', 'happy', 'neutral', 'sad', and 'surprise'.

Data Source

The model is trained on the FER dataset from Kaggle, containing various facial expressions categorized into seven classes. For more information and access to the dataset, visit the FER2013 Dataset on Kaggle.

Dataset Composition

Training Set:

Emotion	Number of Images
Angry	3993 images
Disgust	436 images
Fear	4103 images
Happy	7164 images
Neutral	4982 images
Sad	4938 images
Surprise	3205 images

Test Set:

Emotion	Number of Images
Angry	960 images
Disgust	111 images
Fear	1018 images
Happy	1825 images
Neutral	1216 images
Sad	1139 images
Surprise	797 images

Model Architecture

This system uses a sophisticated machine learning model built on the ResNet50V2 architecture, pre-trained on ImageNet and adapted for emotion detection. It includes features such as Batch Normalization, Global Average Pooling2D, and Dense layers with Dropout, culminating in a softmax activation function for classifying the seven emotions.

Performance Metrics

• Training Accuracy: Improved from 21.2% to 63.3% over 50 epochs.
• Validation Accuracy: Increased from 23.7% to 66.6%.

Some Results

Videos

Install from Source

# Create a Conda environment
conda create --name facial_emotion_recognition python=3.10
# Activate the Conda environment
conda activate facial_emotion_recognition

# Clone the repository
git clone https://github.com/Rizwanali324/Facial_Recognation-Emotions_Classification.git
cd facial_emotion_recognition

# Install the project dependencies
pip install -r requirements.txt
**Running the Code**
python code/emotion.py


**Converting the Model to TFLite**

python code/converter.py

Tech used

Future Enhancements

To enhance model performance and robustness:

• Dataset Expansion: Include a broader range of facial expressions and real-world conditions.
• Multimodal Data Integration: Incorporate audio cues alongside visual data.
• Advanced Neural Architectures: Explore deep convolutional networks, recurrent neural networks, or attention-based models.

Conclusion

This emotion detection project aims to enhance human-computer interaction by enabling computers to interpret human emotions accurately, demonstrating promising applications in various real-world settings.