Determining appropriate no. of clusters..

16 views
Skip to first unread message

geeta sharma

unread,
Apr 12, 2012, 11:49:08 PM4/12/12
to face...@googlegroups.com
While clustering face images by kmeans function how can we determine appropriate value of k (i.e. appropriate no. of clusters) for efficient clustering?

Kourosh Meshgi

unread,
Apr 13, 2012, 12:27:30 AM4/13/12
to face...@googlegroups.com
use cross validation, grow K from 2 to N and find the trend in validation results, that will give you the optimal K for that particular clustering...

On Fri, Apr 13, 2012 at 12:49 PM, geeta sharma <sharm...@gmail.com> wrote:
While clustering face images by kmeans function how can we determine appropriate value of k (i.e. appropriate no. of clusters) for efficient clustering?

--
You received this message because you are subscribed to the Google Groups "Face Recognition Research Community" group.
To post to this group, send email to face...@googlegroups.com.
To unsubscribe from this group, send email to face-rec+u...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/face-rec?hl=en.

sabrina

unread,
Apr 19, 2012, 2:50:26 AM4/19/12
to Face Recognition Research Community
It hard to determine appropriate value of k using kmeans.
Why not choose other algorithms of cluster?

Kourosh Meshgi

unread,
Apr 19, 2012, 3:01:52 AM4/19/12
to face...@googlegroups.com
It is not hard, it is more trustable algorithm due to it's simplicity...
And for the case of Haar filter:  http://www.tomgibara.com/computer-vision/haar-wavelet 


2012/4/19 sabrina <hhu...@gmail.com>
Reply all
Reply to author
Forward
0 new messages