Hello elham h, I think LBPH, SIFT or SURF could be good options but it all depends on your dataset. In my opinion, you should test these algorithms in your dataset and see which performs better. You can do it “easily” using the implementation provided by the OpenCV library. One tip to improve the average of positive facial recognition would be to treat the illumination variation, the background, and even the low resolution (if possible) before performing the recognition. You can find more posts about face recognition and image processing in this amazing blog www.pyimagesearch.com.
I hope it helps.
Regards