c++ - What is a good training dataset for face recognition? -

c++ - What is a good training dataset for face recognition? -

i working on recognising single person out of set of 20 individuals. currently, software detects faces, low-pass filtering cut down false positives, , records cropped faces folder.

this documentation on cascade classifier training great. however, can't address specifics of face recognition.

negative images

negative samples taken arbitrary images. these images must not contain detected objects.

... each image should (but not nessesarily) larger training window size

...placing [the positive sample] on arbitrary background

what best negative samples face recognition?

other faces or objects? close expected lightning , face orientation or random? how determine number of samples (as samples need prepared hand, if faces)? larger positive images or same size? or need many shots of possible backgrounds? positive images

you may need 1 positive sample absolutely stiff object opencv logo, definetely need hundreds , thousands of positive samples faces. in case of faces should consider race , age groups, emotions , perhaps beard styles.

the above extract aimed @ face detection, big sample. need observe single person. sane number of positive samples application?

c++ opencv