Confidence in OpenCV FaceRecognizer predict method output

Question

I am using OpenCV for face recognition and have a newbie question. Here is a portion of my code:

recognizer = cv2.createLBPHFaceRecognizer()
...
nbr_predicted, confidence = recognizer.predict(predict_image)
...

My question is the higher the confidence is means faces are more similar or less similar?

Answer 1

If the confidence is higher, then it means that the pictures are less similar, or in other words the lower, the better. Many people, including me, found this weird but I just changed my variable to be called loss instead of conf or confidence. Hope that answers your question.

Answer 2

There is an implementation about face recognition wich you can read here. They use OpenCV with face module.

In their read me there is an explanation about project and they say that confidence was more similiar when the number is lower. And vice-versa.

I have been studying about cv2.face with eigenfaces and fisherfaces and I received numbers around 10000 in my predictions but with LBPH I could predict with distance of 60~80.

Do you have any project about LBPH to share?