# FaceRecognizer Confidence

I'm developing an OpenCV Face Recognition application in C++ in VisualStudio 2010. I've been using all three methods, but mostly leaning towards LBP. The point of the application is to compare percent similarity of images of faces to a specific face from a training set. Is there a way to turn the returned confidence value from the predict method into a percent value that makes sense? I know that these values correspond somewhat to a distance between faces, and a lower number corresponds (inversely) to a higher confidence level, but where does the actual number come from? I want to return a percent similarity for my application, but I'm not sure what the returned confidence means. Specifically, does anyone know anything about the confidence value pertaining to the LBP method? Does anyone know what the range of possible values for confidence are?

Thanks a bunch!

Natalie

edit retag close merge delete

Sort by ยป oldest newest most voted

more

Thanks for the response, but I really need a range of possible distances, or a more specific formula for how it's computed for LBPH. I need to be able to turn the value of confidence into a percent, and having that information is the only way I can accomplish that.

( 2014-03-27 15:09:24 -0500 )edit

Yes, I think you need to determine it empirically using data that is relevant to your application. There is no built-in way of converting the distance metric to a percent. If you feed large numbers of relevant examples into the recognizer and keep all the returned metrics, you can do statistics on that.

( 2014-03-27 16:19:05 -0500 )edit

Gotcha. So you're saying, if I think 200.00 is an appropriate confidence threshold for my application, I should threshold it there, and treat 0 as a max and 200 as a min, and create a percent out of that? Sorry if that doesn't make sense, haha. Thanks for your help!

( 2014-03-27 16:30:23 -0500 )edit

Something like that :-)

( 2014-03-28 13:51:02 -0500 )edit

Official site

GitHub

Wiki

Documentation