CODEWITHSUNDEEP
c++opencv
Michael Chourdakis
Michael Chourdakis

Reputation: 11158

OpenCV FaceRecognition issue - it almost always matches images

I am working on a Face Recognition project, using OpenCV.

I am training all the 3 recognizers (Eigen,Fisher, LPBH) using the CvHaarClassifierCascade. The problem is that FaceRecognition::predict almost always predicts a match.

For example, let's say I have 10 images per person and I have 2 persons. When person #1 sees the camera, prediction works. When person #2 sees the camera, prediction also works.

However when I have only 1 person, then when person #2 sees the camera, LPBH almost always matches it to person #1. That is, when the person looking the camera is not included in the training classes, LPBH almost always matches him.

Are there parameters I could configure the LPBH with in order to avoid this behaviour ? I want it to reject "unknown" persons.

Best Regards.

Upvotes: 1

Views: 234

Answers (1)

Tõnu Samuel
Tõnu Samuel

Reputation: 2896

Classifier can be trained to make difference between TRAINED examples. If you want to reject samples which look somewhat different, check for confidence level. Classifier basically tells "I think, this is person #1 but my confidence is 0.39 from possible maximum 1.0".

Upvotes: 1

Related Questions