Introduction - If you have any usage issues, please Google them yourself
The neural network s input can be lower resolution of face image, local area of the autocorrelation function, local texture of second order moment, etc. This kind of method also need more training samples, and in many applications, the sample size is very limited.