I created a biometric model for facerecognition using OpenCV’s LBPH algorithm (OpenCV: Face Recognition with OpenCV).
Anyone know if it is possible to calculate evaluation metrics like FAR, FRR and ERR?
hmm, may i ask, what you are trying to achieve here ?
what is the exact purpose of your face recognition ?
opencv’s FaceRecognizer classes are doing “identification”
(that is, they return the closest item from a large database),
while terms like FAR and FRR only make sense in a “authentification” (is that me ?) or “verification” (same person as in the other image ?) context
FAR and FRR are nicely explained here, but, there is nothing like an “imposter” or a “rejection” here, just correct or incorrect predictions
however, as long as you try to use that code, you should
build a “confusion matrix”, (for 10 persons, a 10x10 matrix, which you increase like
conf[predicted, ground_truth] ++ for each prediction in your dataset), which will give you:
- correctly identified persons on the diagonal
- incorrectly identified as paul on the resp. row
- paul incorrectly identified as someone else on the resp. col