FaceRecognition

A lightweight, software-only face recognition system using OpenCV's Local Binary Patterns Histograms (LBPH). This project allows you to capture face data, train a local model, and run real-time recognition via a desktop webcam.

Note: This repository was adapted from an Arduino-integrated project. All hardware dependencies have been removed for a pure software implementation.

---

📂Repository Structure

Face recognition
├── app.py                # Real-time recognition (uses trainer.yml & labels.json)
├── capture_faces.py      # Captures labeled face images to dataset/
├── train_model.py        # Trains LBPH model; generates trainer.yml & labels.json
├── camera_test.py        # Diagnostic tool for webcam access
├── dataset/              # Created by capture_faces.py (contains person subfolders)
├── trainer.yml           # Trained model file 
└── labels.json           # ID-to-Name mapping (created during training)

---

🛠️ Requirements & Installation

Python: 3.8+ (3.10+ recommended)

Environment: Virtual environment recommended

1. Setup Environment

python -m venv .venv

# Windows
.venv\Scripts\activate
# macOS / Linux
source .venv/bin/activate

2. Install Dependencies

pip install --upgrade pip
pip install opencv-python opencv-contrib-python numpy

Note: opencv-contrib-python is strictly required for the LBPH modules.

---

🚀 Basic Usage

Follow these three steps in order to get the system running:

Step 1: Capture Labeled Faces

Run the capture script to build your dataset.

python capture_faces.py

• Input: Enter the name of the person when prompted.
• Samples: Default is 50. For better accuracy, aim for 200–500 images.
• Controls: Press Enter to finish early or q to cancel.

Step 2: Train the Model

Process the images in the dataset/ folder into a machine-readable model.

python train_model.py

This assigns numeric IDs to folders and generates trainer.yml and labels.json.

Step 3: Run Recognition

Start the live desktop demo.

python app.py

• The script uses your webcam to identify faces in real-time.
• Controls: Press q to exit the window.

---

📈 Improving Accuracy

• Sample Size: Capture 500+ images per person.
• Variety: Capture faces under different lighting, different angles, and with/without glasses.
• Thresholding: Adjust the CONFIDENCE_THRESHOLD in app.py. A lower number makes the system "stricter" about who it recognizes.

---

🔧 Troubleshooting

Issue	Solution
trainer.yml not found	Run train_model.py first. Ensure dataset/ is not empty.
Camera not opening	Change the index in cv2.VideoCapture(0) to 1 or 2.
AttributeError: 'cv2' has no 'face'	Install opencv-contrib-python.
Low Accuracy	Increase dataset size, improve lighting, and ensure faces are cropped/resized properly.

---

🔒 Privacy & Safety

• Local Storage: All face data and models are stored locally on your machine.
• Consent: Do not capture or store face data of individuals without their explicit consent.
• Security: If using in a production environment, ensure trainer.yml and labels.json are stored securely.

---

📜 License

This project is licensed under the MIT License.

---

📸 Screenshots of Output

• facial recognition of existing datasets:

• setting up a new facial recognition :