Quote:
but for any reason ,if i have 1000 trained face ,i think it would be take long time to compare it ,
It depends on what ML algorightm you are using. It's not a pixel by pixel comparison that you do to identify the person and label them. Once ML is trained, giving a new face as input will give back closest match. You can cap a percentage on closeness as same or different.
Now, to have some thoughts, have a look at this for more details:
EMGU Multiple Face Recognition using PCA and Parallel Optimisation[
^]