I've been using OpenCV on a small production application to do face recognition with children of ages 1 to 5. For the face recognition and subject matching we are using a LBPH face recognizer with standard parameters. Most of the time we get positive results, but with a sample of 100 children we get roughly a 65% of succesful matches. For the recognition model we use a total of 5 images per child.
Is there are any advice you could give to improve the face matching? Would Eigenfaces or Fishfaces be better suited for the recognition?
Any help will be extremely appreciated.