facial landmarks detector training (MATLAB)
This project is used to train landmark detectors using linear SVM and logistic regression. The final outcome is the batch of linear classifiers for 88 landmarks of human faces.
Given training samples and landmark annotations, features are extracted in PrepareTrainingData.m. Positive samples are extracted from annotated position, while negative samples are sampled within a constant shift from positive ones. Annotation data are loaded as .mat file first before usage.
Local Binary Pattern image is extracted by lbp.m. Linear SVMs are trained in SVM_Train.m. Finally, logistic.m performs logistic regression on results of linear SVMs. LandmarkDetectorTrain.m combines all the procedures into a whole. init_train_script.m is the Matlab script file managing data I/O and alogrithm. Incoporate your own data by modifying this file.
If you have any problems, please contact me via [email protected].