Compare two faces and return the similarity percentage

asked 2015-02-23 01:51:08 -0500

SHIVA gravatar image

I want to write a program (perhaps python script) which can take two human face images as input and should

compare the faces return the similarity % or matching % between given images,

same like This

edit retag flag offensive close merge delete



sorry, but this is not an easy task, and will require serious knowledge about computer-vision.

have a look here , to see, what you're up to.

berak gravatar imageberak ( 2015-02-23 01:59:46 -0500 )edit

IMHO it's based on some features, and matching, but I cannot figure out how the percentage is computed... maybe based on how many features were matched? @berak: Do you think it has a face recognizer behind?

thdrksdfthmn gravatar imagethdrksdfthmn ( 2015-02-23 06:46:27 -0500 )edit

^^ no, i don't think so.

the difference between face-identification (that's what opencv's face-reco does) and face-verification (above task) is, that you have a lot of known faces in the 1st case, and just find the closest, while above , you only got those 2 images, and can't ever expect to have another image of either person at hand.

what you have to do here, is to learn a distance or 'metric', that can distinguish same vs not same, trained on many same/not-same image pairs. unfortunately, finding good features and classification methods for this is like 10 times harder than in the identification task (where you can just use more images of the same person)

@SHIVA, the estimate to build something like that from scratch without prior knowledge is ~1 year (of solid work).

berak gravatar imageberak ( 2015-02-23 07:10:38 -0500 )edit