Image classification using Trainable Weka Segmentation

[luckyluke]

I am using Trainable Weka segmentation on an sample image in ImageJ (Dot_blog.jpg)

I just made some classes and assign some dots to different classes. When I generate the output in  arff format. there are 1721 samples corresponding to default setting by Weka segmentation


The figure is 513*282 *8 bits. Can someone explain why there are 1721instances   . I exactly don't understand the segmentation in my arff file?

I actually want to do an image classification based on a series of traning images (image sequences) so each (whole) image is classified wtihin one class but I am not sure exactly what weka segmentation is doing?

Thanks