2

I need a collection of sample images to train a Haar-based classifier for face detection. I read that a ratio of 2 negative examples for each positive example is acceptable. I searched around the web and found many databases containing positive examples to train my classifier (that is, images that contain faces), however, I can't any database with negative examples. Right now, I have over 18000 positive examples. Where can I find 2000 or more negative examples?

Douglas De Rizzo Meneghetti
  • 916
  • 1
  • 8
  • 26
  • Could you please try and fix the spelling/grammar issues. As of now, your post is unreadable. – Lilith Daemon Sep 06 '15 at 23:15
  • which word you doesn't understand it ? –  Sep 06 '15 at 23:38
  • The entire piece, you fixed some spelling errors, but it still is really confusing. Are you asking for how to extract the background? How to find a database of images? How the haar training works to read an image, and how you can generate sample images? Any of these are possible interpretations of your question. – Lilith Daemon Sep 06 '15 at 23:46
  • no I ask where can i find negative image more than 20000 downloading it from google take long time –  Sep 06 '15 at 23:48
  • just use any images where the object-to-detect isnt present. You can use rotations/scales and subimages of those negative samples too. – Micka Sep 07 '15 at 07:00
  • i know any image but i need more than 20000 any way for download 20000 image from google street ? –  Sep 07 '15 at 21:52
  • I guess you could take a lot of screenshots... Honestly, without more information I don't know what else to tell you. What you are asking for really is vague. (Maybe checkout MIT, Carnegie, etc. They often have databases of images. [Although not likely 20k, that is a REALLY big amount]. ) You might have to generate them yourself if you need that many. – Lilith Daemon Sep 07 '15 at 23:37
  • The link you provided has roughly 5k negative images, what are you trying to train it to do that you need much more than that? – Lilith Daemon Sep 07 '15 at 23:38
  • i have 15000 positive image so i must have more than 15000 for negative for this tow days i have now 8000 –  Sep 08 '15 at 15:48
  • You still haven't said what "object" you are training it to find. How are we supposed to suggest the correct database without this information. – Lilith Daemon Sep 09 '15 at 12:51
  • i say its face frontal and profile –  Sep 09 '15 at 13:31
  • now i just test on 1200 positive image when its on stage 7 its detect some faces with some false detect but when its be in stage 13 its dosn't detect any thing why ? –  Sep 09 '15 at 13:33

1 Answers1

0

use http://tutorial-haartraining.googlecode.com/svn/trunk/data/negatives/

or any other image set that has no objects you need to recognize

NOTE: the number of samples you mention is too big, you don't need so much to obtain high accuracy

Yavpolnom Shoke
  • 78
  • 10
  • i want to use 30000 image for face detection did thats too big how much must i use ? for best detection left side and frontal face –  Oct 07 '15 at 17:07
  • 1
    the original cascade in openCV uses 5000 positive samples - but for front faces only. in your case it might be increased or split in two cascades. which is a better approach - better to check on real data. – Yavpolnom Shoke Oct 08 '15 at 01:40