🎭 Face Mesh Processor

This repository demonstrates real-time facial landmark detection using MediaPipe Face Mesh. It allows processing:

• Static images
• Video files
• Live webcam feed

The project uses MediaPipe, a Google library for cross-platform, real-time face, hand, and body tracking.

Link : Official MediaPipe Face Mesh documentation

🧩 Repository Structure

├── README.md
├──requirements.txt       
├── facemesh_processor.py # Shared processor class for face landmarks
├── image_process.py      # Script to process a single image
├── video_process.py      # Script to process a video file
├── webcam_process.py     # Script to process webcam feed
├── extractLandmarks.py   # Landmarks Extractor from face 
├── landmarks2txt.py      # Landmarks → txt
├── exemple_image.jpg/    # Example input image
└── exemple_video.mp4/    # Example input video

📦 Installation

Ensure Python 3.7+ is installed. Install the dependencies:

pip install opencv-python mediapipe 

🧠 How It Works

1️⃣ Face Mesh Processor

The FaceMeshProcessor class contains the main logic:

• Converts images to RGB format (required by MediaPipe).

• Detects facial landmarks using mp.solutions.face_mesh.

• Draws landmarks (tessellation, contours, and irises) on images.

• Returns processed images for display or saving.

2️⃣ Image Processing

• Reads a single image.

• Uses FaceMeshProcessor to detect and draw landmarks.

• Saves the annotated image.

python image_process.py

3️⃣ Video Processing

• Reads video frame by frame.

• Uses FaceMeshProcessor for each frame.

• Writes processed frames to a new video file.

python video_process.py

4️⃣ Webcam Processing

• Captures live feed from the webcam.

• Processes each frame in real-time.

• Displays annotated frames.

python webcam_process.py

🕹️ Interest of Face Landmarks

1️⃣ extractLandmarks.py

Purpose:

• Extract 3D facial landmarks from images using Mediapipe Face Mesh.

• Provides a convenient class FaceLandmarkExtractor that returns the landmarks in a flattened array format.

• Can be used for single images or video frames.

Key Functions:

class FaceLandmarkExtractor:
    def __init__(...):
        # Initialize Mediapipe Face Mesh
        pass

    def extract(self, image, normalize=True):
        # Extract facial landmarks from an image
        # Returns a flat 1D list: [x0, y0, z0, x1, y1, z1, ...]
        # If normalize=True, coordinates are shifted to start from 0
        pass

How to use:

from extractLandmarks import FaceLandmarkExtractor
import cv2

extractor = FaceLandmarkExtractor()

image = cv2.imread("happy_face.jpg")
landmarks = extractor.extract(image)  # landmarks is a list of 1404 floats (468 points x 3)

→ Landmarks can now be used as input features for an ML model.

2️⃣ landmarks2txt.py

Purpose:

• Loops through a dataset of images organized by emotion folders.

• Uses FaceLandmarkExtractor to extract landmarks for all images.

• Adds the emotion label to the end of each landmark vector.

• Saves the result as a text file data.txt for training ML models.

Expected folder structure:

data/
    happy/
        img1.jpg
        img2.jpg
    sad/
        img1.jpg
        img2.jpg

• Each row in data.txt has 1405 values → 1404 landmarks + 1 label.

3️⃣ Workflow to Train Emotion Model

1. Collect dataset of faces organized by emotion.

2. Run landmarks2txt.py → generates data.txt.

3. Load data.txt → features = landmarks, labels = emotion.

4. Train ML model (Dense NN or other) on this data.

For live prediction:

Capture frame → extract landmarks → feed to model → get emotion.

📝 Notes

• Supported image formats: .jpg .png

• Supported video formats: .mp4 .avi

Make sure you replace input.jpg and input.mp4 with your own files.