Details are on ausome.tistory.com (Korean)
You should gather the training data by yourself. I gathered the training data from open sources as like CMU, Jaffe, Yale, etc and from google image, my friends' face.
There are some preprocessing steps. 1. crop the face region, 2. perform the Homomorphic filtering, 2. blurring and histogram equalization.
I made the training data path into a text file and put it into the input. If you have better way to put in the input data, you can feed the data in your own way.
=====================================================================
Alexandra.cpp file is for I.MX6Q board