Emotion Detection from Images using CNN

📌 Project Overview

This project focuses on detecting human emotions from facial images using Convolutional Neural Networks (CNN). Trained on the FER-2013 dataset, the system can classify seven basic emotions: Angry, Disgust, Fear, Happy, Sad, Surprise, Neutral.

🧑‍💻 Group Information

• Group No: 8
• Group Name: leisurely_loco

👥 Team Members

• Arbin Zaman (ID: 2125051006)
• Sohana Afrin (ID: 2125051013)
• Safin Ahamed Sajid (ID: 2125051022)
• Md. Fahim Abrar Asif (ID: 2125051116)

🎯 Objectives

• Build a CNN model to recognize facial emotions
• Use the FER-2013 dataset for training and validation
• Enable emotion detection in real-time from static images or webcam
• Evaluate model performance and optimize for accuracy and generalization

⚙️ Tools and Environment

• Language: Python
• IDE: Google Colab
• Libraries: TensorFlow/Keras, OpenCV, Matplotlib, NumPy
• Dataset: FER-2013 (from Kaggle)

🔧 Setup Instructions

1. Clone the repository:

   git clone https://github.com/arbinzaman/Emotion-Detection-System.git
   cd Emotion-Detection-System

2. Install required libraries:

   pip install tensorflow opencv-python matplotlib numpy

3. Download FER-2013 dataset:

   • Register and accept terms on Kaggle.
   • Use kaggle datasets download -d msambare/fer2013 or manually upload the dataset to Colab.

🧠 Model Architecture

The CNN consists of:

• 2 convolutional layers with ReLU activation and max pooling
• Dropout layers to reduce overfitting
• Dense layers with softmax for emotion classification (7 classes)

🚀 How to Run

Run the Jupyter Notebook (emotion_detection_colab.ipynb) step-by-step:

• Preprocess dataset
• Train CNN model
• Use predict_emotion(img_path) function to test images

📸 Sample Output

Input Image	Predicted Emotion
	😄 Happy

📈 Results

• Validation Accuracy: ~60%
• Real-time predictions on unseen images work reliably under various lighting conditions
• Robust against moderate noise and variations

🧪 Limitations

• Struggles with underrepresented emotions like "Disgust" and "Fear"
• Grayscale images only – lacks color cues
• Sensitive to extreme lighting or occluded faces

🔮 Future Work

• Use colored, high-resolution images
• Add real-time webcam support
• Deploy via TensorFlow Lite for mobile apps
• Integrate multimodal data (voice/text)

📚 References

1. FER-2013 Dataset on Kaggle
2. François Chollet, Deep Learning with Python, Manning, 2017
3. TensorFlow Keras API
4. OpenCV Documentation

🔗 GitHub Repository

https://github.com/arbinzaman/Emotion-Detection-System