Face Detection Parameters

asked 2017-05-03 13:07:36 -0600

updated 2017-05-04 05:24:04 -0600

berak
32993 ●7 ●81 ●312

Hello, am currently working with a face detection - recognition system, and currently am in the testing stage, but i am confused with Detection phase I have searched a lot, and came to many important points.

The detection i.e.

face_cascade.detectMultiScale(gray, 1.05, 3,minSize=(20, 20), maxSize=(200,200))

how should i choose, scaling factor, and how is it important (i didn't totally understood what its useful for though i read a lot ), and how to choose min. neighbors and the window size to ensure that i will be able later on to recognize these faces.

answered 2017-05-04 05:19:49 -0600

berak
32993 ●7 ●81 ●312

updated 2017-05-04 05:20:38 -0600

scale factor. "MultiScale" means, an image pyramid is searched, where each level is a factor of X larger (1.05 here) smaller values will make a more exact detection, larger values will make it faster.
minNeighbours. this determines, how many detections (at different scales) have to be found, to make it a valid detection. increase it, if you get false positives, decrease it, if you start to miss faces.
minSize. obviously, the minimum size to search for faces. now, since all the cascades were tryined on 24x24 rects, and since it can never detect something smaller than that, you probably should increase it to 30x30 or such.
maxSize. if left empty, it will search up to the whole image size.

edit flag offensive delete link

Comments

Thanks a lot, thats was really helpful, one more thing, is theres any suggestions about the best values for theses parameters to choose ?

Ameeer ( 2017-05-07 15:37:43 -0600 )edit

add a comment

Face Detection Parameters

1 answer

Comments

Links

Question Tools

Stats

Related questions

Face Detection Parameters edit

1 answer

Comments

Links

Question Tools

Stats

Related questions

Face Detection Parameters