Selecting suitable negative aka background image for haartraining is very important. Collect set of images (about 3000) that does not contain objects of interest, for an example if you are going to train a haarcascade classifier for Face, you should have negatives with no face images.
you can download collection of negative images (about 3000 negatives) from here. [Use Internet Download Maneger]
http://tutorial-haartraining.googlecode.com/svn/trunk/data/negatives/
According to your haartraining requirement this may not be not enough . best proposition for No. Positive images : No. Negative images is 1:2, means if you have 2000 positive images you should have at least 4000 negative images. So if this is not enough, you can get more from public databases such as google images.
You can use these softwares to make download easy otherwise it would be boring . below softwares would be helpful
[1]bulk image downloader (commercial software)
[2]google image dowloader (FREE software) http://www.googleimagedownloader.com
After you get enough images you should do index and rename all images properly for further use.This tool would be useful
Ant Renamer ( http://www.antp.be/software/renamer )
Related Post
[ Opencv installation ]
http://www.tectute.com/2011/01/opencv-22-installation-on-visual-studio.html
[ Positive images for haartraining ]
http://www.tectute.com/2011/06/haartraining-positive-samples.html
[ Opencv haartraining and haar cascade xml ]
http://www.tectute.com/2011/06/opencv-haartraining.html
This link is not available now:
ReplyDeleteLink: http://tutorial-haartraining.googlecode.com/svn/trunk/data/negatives/