You may want to read these two papers:
"A Meta-Analysis of Face Recognition Covariates", Yui Man Lui, David
Bolme, Bruce A. Draper, J. Ross Beveridge, Geoff Givens, P. Jonathon
Phillips
and "Meta-Analysis of Face Recognition Algorithms", P. Jonathon
Phillips, Elaine M. Newton to learn more about standard methods.
Also section 1.1.2 of "Face Processing" book by wenyi zhao and rama
chellappa might give you good idea and help you in this matter. It's
very short though.
good luck
ImAn
On Nov 3, 5:56 am, Alexander Goncharov <ag.ts...@gmail.com> wrote:
> Hello!
> Could somebody point me to the good article about quality assessment
> of face recognition system?
> Denote F = {(f_k,label_k), k=1,...,N} a sample of faces, where each
> face f_k associated with person's id label_k.
> Denote A:F->{1,...,K} a face recognition algorithm, which for each
> face f from F give the person's id. It is obvious, that in some cases
> face recognition algorithm may give wrong id: A(f_k) is not label_k.
> The question is how to estimate Type 1 Error and Type 2 Error (or
> recall and precision) if sample F contain different number of photos
> for each person?
> I know how to estimate this metrics for separate person, but i'm not
> sure how to correctly aggregate this metrics for all persons in sample.
|
