1

I'm quite new to face recognition algorithm, I just found that if I train a simple face database. For example, I have four people (A,B,C,D), and each people has 10 images, then I can train the database very quickly by existing algorithm (such as by OpenCV library). And the face recognition algorithm works quite well. The accuracy is larger than 90%, I can recognize the four people correctly(give images of either A,B,C,D), also I can reject some unknown people(such as people E,F,G...).

But this is a very small face database, I want to know that if I have to recognize 10000 subjects, that may be 100000 images, it is still possible? what is the accuracy in such case?

Basically, face recognition algorithm just does some classification on the feature spaces of the collected faces, but does it still work quite well when the database becomes quite large?

Thanks.

ollydbg23
  • 1,032
  • 1
  • 10
  • 31

1 Answers1

2

Yes, if you want to recognize 10000 subjects, you have MINIMALLY need 10000 images for each of them. Generally , The recognizer algorithm works under distance comparaison among samples of its database, therefor, if the algorithm is very robust, you can easily recognize many subjects you want. However, 10000 is a very big number, you really need a GPU conception for a real time application, cause it take a long time to give you an answer. Hope it help.

Kacem
  • 116
  • 6
  • Thanks for the answer! I will try to find some robust face recognition algorithms, it looks like there is a page asking such question: [Face recognition Library](http://stackoverflow.com/questions/953714/face-recognition-library) – ollydbg23 Apr 02 '15 at 21:48