🎭 Facial Emotion Recognition System

Real-time detection and classification of 7 human emotions from a live webcam feed using Haar Cascade face detection and a pre-trained CNN via DeepFace.

---

📸 Demo

Webcam opens → face is detected → emotion label + confidence bars appear in real time.

😊 Happy  72%   ████████████░░░░
😐 Neutral 15%  ███░░░░░░░░░░░░░
😢 Sad      5%  █░░░░░░░░░░░░░░░
😠 Angry    3%  ░░░░░░░░░░░░░░░░
😱 Surprise 3%  ░░░░░░░░░░░░░░░░
😨 Fear     1%  ░░░░░░░░░░░░░░░░
🤢 Disgust  1%  ░░░░░░░░░░░░░░░░

---

✨ Features

• 🎥 Live webcam — real-time processing frame by frame
• 🔍 Haar Cascade face detector — fast, CPU-friendly frontal face detection
• 🧠 DeepFace CNN — pre-trained on FER-2013 (35,887 labelled face images)
• 📊 Emotion bar chart — all 7 scores displayed alongside the face
• 🔄 Smoothing — averages last 5 results to prevent flickering
• ⚡ Throttled analysis — DeepFace runs every 0.4s to keep CPU usage low
• ⌨️ Keyboard controls — quit, screenshot, reset smoothing

---

🧠 Emotions Detected

Based on Paul Ekman's 7 Universal Emotions (1972):

Emotion	Label
😊	Happy
😢	Sad
😠	Angry
😨	Fear
😱	Surprise
🤢	Disgust
😐	Neutral

---

🗂️ Project Structure

facial-emotion-recognition/
│
├── emotion_detection.py     # Main script — webcam loop + detection + drawing
├── requirements.txt         # Python dependencies
└── README.md                # You are here

---

⚙️ Tech Stack

Tool	Purpose
Python 3.8+	Core language
OpenCV (cv2)	Webcam capture, image drawing, Haar Cascade
DeepFace	Pre-trained emotion CNN (FER-2013)
NumPy	Score smoothing & array operations
Haar Cascade XML	Fast frontal face detection (built into OpenCV)

---

🚀 Getting Started

1. Clone the repository

git clone https://github.com/salonisharma2004/facial-emotion-recognition.git
cd facial-emotion-recognition

2. Install dependencies

pip install -r requirements.txt

3. Run

python emotion_detection.py

Note: DeepFace will auto-download the emotion model (~80MB) on first run. Make sure you have an internet connection.

---

⌨️ Controls

Key	Action
Q	Quit the application
S	Save a screenshot
R	Reset smoothing history

---

🔧 How It Works

Webcam Frame
     │
     ▼
Convert to Greyscale + Histogram Equalisation
     │
     ▼
Haar Cascade → detect face bounding boxes (x, y, w, h)
     │
     ▼
Crop face ROI → pass to DeepFace.analyze()  (every 0.4s)
     │
     ▼
CNN returns scores for 7 emotions  (softmax probabilities)
     │
     ▼
Average last 5 results (smoothing)
     │
     ▼
Draw bounding box + label + bar chart on frame
     │
     ▼
cv2.imshow() → display to screen

---

📦 Requirements

opencv-python
deepface
tf-keras
numpy

Install all with:

pip install -r requirements.txt

---

🛠️ Troubleshooting

Problem	Fix
Webcam not opening	Change cv2.VideoCapture(0) to (1) or (2)
Too many false face detections	Increase minNeighbors from 5 to 7
Missing faces	Decrease scaleFactor from 1.1 to 1.05
Slow performance	Lower resolution: set width to 640, height to 480
DeepFace model not downloading	Check internet connection on first run

---

🌱 Future Improvements

• Add support for image/video file input (not just webcam)
• Export emotion logs to CSV
• Build a Flask web app for browser-based demo
• Train a custom CNN on a larger dataset
• Add multi-face tracking with unique IDs

---

👩‍💻 Author

Saloni Sharma CS Student | Python | ML | Computer Vision GitHub

---

📄 License

This project is licensed under the MIT License — feel free to use and modify it.