Designer: Junbo Zhao, homepage
This project focuses mainly on facial expression recognition, which has been widely used in varies of fields. Take ITS(Intelligent Transportation System) as an example, a good tool of facial analysis is able to judge the physical and pshychological condition of the driver and further reduce the possibility of hazard.
My code is mainly comprised of three parts:
1. Find the positions of facial landmarks, including eyes, nose and mouth.
2. Extract LBP features over landmarks in a multi-scale and multi-cell manner. We normally ignore information from other parts of faces except the landmarks and their surroundings, since other parts would not bring us a discriminative classifier.
3. Calculate the difference between the face of a special expression and the neutral courterpart. Consequently the difference, namely a long vector, is that expression's feature.
4. Prepare lots of features following the steps above and Train a SVM.
5. Use the trained model to test a new face of an unknown expression.
To be updated!
Ubuntu 12.04 LTS, g++ 4.6.3, OpenCV 2.4.7