I'm working on palm-print identification system and I want to evaluate my system, i read in a couple of papers the researchers using GAR, FAR, FRR, EER and ROC, the question is that suppose I have the following confusion matrix which resulted from a classifier(example):
13 **** 2 **** 5
0 **** 20 **** 0
3 **** 12 **** 5
How i can calculate those values from this confusion matrix and plot them as a curves or graph ( see the attached image )
one more thing, are recognition and identification have the same meaning in bio-metrics?
FAR, FRR and then ROC curves are all defined for 2 classes problems only, and not multi classes as you have, i.e. authentication not identification
However u may do an ROC curve for each subject considering 2 classes : the subject itself (class a) and all the other classes are to be (class b). You may do this for each class then you may average all.
And to get different FAR/FRR to draw the curve u need to change the matching threshold if u have any and record both of FAR and FRR.
So it is a long process and might not mean anything for identification so we normally do not do them as we have tones of other problems to worry about, it is enough to record precision, recall and accuracy.
recognition and identification are sometimes used interchangeably, however, recognition is more general term, and might refer to ether identification or authentication, so it is better to be more specific and use identification or authentication.