FaceMesh (AI WITHOUT ML)

Project Overview

This project demonstrates real-time facial landmark detection using MediaPipe Face Mesh.

Note: Multiple people supported.

The project allows you to:

• Detect and visualize facial landmarks in real-time.
• Highlight eyes, lips, nose, nostrils, and iris with different colors.
• Use Python and OpenCV for webcam capture and image preprocessing.

Recommended Python version: 3.12.7 (MediaPipe has compatibility issues with Python 3.13+).

---

Project Structure

project/ │── 01_create_dataset.py │── 02_review_dataset.py │── 03_train_model.py │── 04_predict.py │── 05_reset_projet.py │── dataset/ → your captured faces │── models/ │ ├── lbph_model.xml │ └── label_map.json │── README.md │── requirements.txt

Technologies Used

• Python 3.12.7
• OpenCV – Webcam capture and image preprocessing.
• MediaPipe – Real-time facial landmark detection.
• NumPy – Array manipulation for image processing.
• protobuf, attrs, matplotlib – Required dependencies for MediaPipe.

---

Installation

1. Clone the repository:

git clone https://github.com/Mchiir/face_recognition_mediapipe_lbph.git
cd face_recognition_mediapipe_lbph

2. Create and activate a virtual environment with Python 3.12.7:

# Windows
"C:\Program Files\Python312\python.exe" -m venv .venv
.venv\Scripts\activate

3. Install the required dependencies:

pip install --upgrade pip
pip install --upgrade --no-deps --force-reinstall -r requirements.txt

Project flow logic

1. Capture Face Images

Run:

python 01_create_dataset.py

You will be asked to enter a name for current person/character. Images are saved to: dataset/<your_name>/

---

2. Review dataset

Run:

python 02_review_dataset.py

Follow onscreen commands to clean your data for model training.

---

3. Train the LBPH Model

Run:

python 03_train_model.py

This will generate: models/lbph_model.xml and models/label_map.json

---

4. Run Face Recognition

Run:

python 04_predict.py

The camera window will show you your face landmark and recognized name.

---

5. Reset project (Optional)

Run:

python 05_reset_project.py

---

Notes

• Always run the script inside the Python 3.12.7 virtual environment to avoid dependency issues.
• Adjust cap = cv2.VideoCapture(1) if your primary webcam is at a different index (0 for default).

License

This project is licensed under the MIT License.