Revision history [back]

The suggestions of Shervin are clear but is impossible for me to have a ideal situation as Shervin has described. What I not understand is why OpenCV makes a lot of errors on recognition. Is possible to have a score (in percentage) with the predict value in order to know how much is reliable the predict value? For me is not important to recognize all the people but is very important that a recognized person is really who he is. Have you got a demo program that I can try in order to know that the errors made from OpenCv are caused from OpenCV and not from a bad OpenCV implementation?