Covariance Matrix Estimation and knn Classifier for Facial Expression Recognition

This program is for facial expression recognition on FaceWarehouse dataset. It can recognize 6 human expressions. There are several algorithms such as Local Binary Pattern (LBP), Multi-scales LBPs known appearance-feature based method and the approach of curvature measurements known as geometric-feature based method implementing in this system.

Solved

• Address many challenges: diverse age, gender, skins, lighting changes even dramatically illumination.
• Interesting discovery: the normal feature plays important role in performance and not yet explain accordingly.

Limitation

• Vary head pose (propose the approach of Iterative Closest Point to rotate the 3D face surface to fixed position )

Installation:

• OpenCV
• The Point Cloud Library
• C++