Active Shape Model Using MUCT Dataset

This project implements an Active Shape Model (ASM) for facial landmark detection using the MUCT face database.

Overview

The Active Shape Model learns facial shape variations from a training set and can generate new face shapes. The implementation includes:

• Mean shape calculation
• Shape alignment using Procrustes Analysis
• PCA-based shape variation modeling
• Visualization of different shape variations

Features

Shape Variations

The model shows three types of shape variations:

1. Mean Shape: The average face shape from training data
2. Component Variations: Shows how faces vary along principal components
3. Random Shapes: Generates possible face shapes by combining variations

Visualization

The project includes visualization tools that show:

• Mean face shape (green points)
• First three component variations (different colors)
• Multiple random shape variations
• Comparisons between different variation types

Installation

bash Clone the repository git clone https://github.com/mabinhang2021/ASM-Using-Muct-Face.git

Usage

(1)Download full MUCT Dataset

1. Visit the MUCT Database
2. Download required files:
   • muct-landmarks-v1.csv (landmark coordinates)
   • muct-*.jpg (face images)
3. Place downloaded files in your project directory: muct-data/ │ ├── muct-landmarks.zip │ └── muct-images.zip

(2)Run the file in ASM/using muct face.py

Results

The model demonstrates:

• Successful alignment of face shapes
• Clear visualization of shape variations
• Effective PCA-based shape modeling

Understanding the Visualizations

The visualization shows three types of shape variations:

1. Mean Shape (Green points)

   • Average face shape from training data
   • Baseline for comparing variations

2. Component Variations (Red, Blue, Yellow points)

   • First three principal components
   • Shows main ways faces vary in dataset

3. Random Shapes (Multiple colors)

   • Different possible face shapes
   • Combines multiple variations

Dependencies

• Python 3.7+
• NumPy
• OpenCV
• Matplotlib
• SciPy
• Scikit-learn

Important Notes

• The MUCT dataset is not fully included in this repository
• You need to download it separately from MUCT Database
• Make sure to place the dataset files in the correct directory structure
• If interested in more information about ASM, please pay attention to this repository, I will add some paper interpretation and pre-mathematical knowledge in the future, thank you

Acknowledgments

• MUCT Database for providing the face dataset