Ask Your Question

Method in opencv for percentile similarity of two faces?

asked 2017-12-11 01:49:27 -0500

zzywd01 gravatar image

Hi guys, is there any method in openCV to get the similarity score in percentage of two faces? I know that SIFT is super to solve this case, but I cannot find any good way to get the percentile result. Could anyone give me some hints? I'm a Java user, but any other language are welcome :)

edit retag flag offensive close merge delete



if you only have 2 images, this is called verification . have a look here

berak gravatar imageberak ( 2017-12-11 01:53:26 -0500 )edit

@berak Thanks for the reply. In fact, I want to compare one test sample with the whole data base, and get the result with the highest percentage.

zzywd01 gravatar imagezzywd01 ( 2017-12-11 02:04:22 -0500 )edit

did u find any solution if yes so kindly share with me i am doing same this project on java/android

vickyrajpoot gravatar imagevickyrajpoot ( 2018-04-09 06:13:00 -0500 )edit

1 answer

Sort by ยป oldest newest most voted

answered 2017-12-11 06:11:11 -0500

@zzywd01, so you need to find

  • distance between your sample and all database samples
  • find the largest distance (worst match)
  • normalize the range [0, largest distance] towards [100%, 0%]

In principle this will always depending on the dataset used, but thats quite normal in my opinion.

edit flag offensive delete link more


And even in my opinion, you could start from [smallest distance, largest distance], because 0 will only be obtained by the exact same image.

StevenPuttemans gravatar imageStevenPuttemans ( 2017-12-11 06:12:44 -0500 )edit

Thank you steven. Is this similar to the value "confidence" which can be obtained from faceRecognizer.predict(test sample, predLabel, Confidence)? It seems that I should normalize the confidence to get the percentage.

zzywd01 gravatar imagezzywd01 ( 2017-12-11 09:28:50 -0500 )edit

Yep thats exactly that I think!

StevenPuttemans gravatar imageStevenPuttemans ( 2017-12-12 08:15:45 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower


Asked: 2017-12-11 01:49:27 -0500

Seen: 523 times

Last updated: Dec 11 '17