search

biometrics / openbr

Open Source Biometrics, Face Recognition
www.openbiometrics.org
Other
2.85k stars 773 forks source link

Retrain FaceRecognition algorithm with impostor score normalization #13

Closed jklontz closed 11 years ago

jklontz commented 11 years ago

Should improve accuracy and give a 0-1 match score range. Also update examples accordingly.

bklare-zz commented 11 years ago

Looking forward to the results.

jklontz commented 11 years ago

Initial results were inconclusive, unfortunately I don't have more time to look into it now to figure out why :(

bklare-zz commented 11 years ago

I believe you said before that you were not using kernel density estimation. If this is true that may be the source of the issue. On Feb 23, 2013 9:05 PM, "jklontz" notifications@github.com wrote:

Initial results were inconclusive, unfortunately I don't have more time to look into it now to figure out why :(

— Reply to this email directly or view it on GitHubhttps://github.com/biometrics/openbr/issues/13#issuecomment-14002016.

jklontz commented 11 years ago

A very plausible explanation, raw distances for this algorithm may very well not be gaussian.

bklare-zz commented 11 years ago

Yeah, they seem to follow more of a single tailed distribution.

Either way, you did put the framework for UNE, which is the thought part with respect to openbr. Hopefully this can be revisited in the future. On Feb 23, 2013 9:49 PM, "jklontz" notifications@github.com wrote:

A very plausible explanation, raw distances for this algorithm may very well not be gaussian.

— Reply to this email directly or view it on GitHubhttps://github.com/biometrics/openbr/issues/13#issuecomment-14002412.

jklontz commented 11 years ago

Not convinced this is worthwhile for us, but adding http://www.ic.unicamp.br/~rocha/pub/papers/2010-eccv-robust-fusion.pdf here for future reference.