Ask Your Question
0

LBPH FaceRecognizer Chi-Square distance

asked 2015-10-02 12:49:58 -0600

WJS gravatar image

updated 2018-01-22 10:33:06 -0600

I’m using the LBPHFaceRecognizer and I understand from reviewing bug 3514 that the correct Chi-Square distance should be CV_COMP_CHISQR_ALT. However, the only Chi-Square distance available in the opencv version 2.4.11 I’m using is CV_COMP_CHISQR. Has anyone compared recognition results for these two Chi-Square distance measures in a controlled way? What prediction errors can I expect if I don’t use the new Chi-Square distance? Finally, is there a way to specify which distance measure to use for the LBPHFaceRecognizer model? WJS

edit retag flag offensive close merge delete

1 answer

Sort by » oldest newest most voted
1

answered 2015-10-05 01:31:50 -0600

berak gravatar image

updated 2015-10-05 01:33:31 -0600

the discussion, which formula is the 'correct' one is probably as old as opencv, they changed it this way and that, and finally came up with having both.. here's some tests on att:

Lbp    H_CHI_ALT    0.973
Lbp    H_CHI       0.945
Lbp    COSINE     0.965
Lbp    HELLINGER   0.981
Lbp    SVM_POLY   0.981
Lbpu  MLP       0.999

unfortunately, atm. there 's no way changing this without hacking the (library)code ;(

edit flag offensive delete link more

Comments

1

Answers @berak, answers ;)

StevenPuttemans gravatar imageStevenPuttemans ( 2015-10-05 06:49:58 -0600 )edit

Berak, I'm not quite clear on what these number mean. Can you explain a little more? Thanks, WJS.

WJS gravatar imageWJS ( 2015-10-08 07:09:21 -0600 )edit
1

accuracy in percent. (for different classifiers on att_faces)

again, get bold, copy the code over to your own project, and hack in any improvements you like.

(then, maybe come back to share)

berak gravatar imageberak ( 2015-10-08 07:15:55 -0600 )edit

Question Tools

1 follower

Stats

Asked: 2015-10-02 12:49:58 -0600

Seen: 383 times

Last updated: Oct 02 '15