| Line 28: | Line 28: | ||
Here we notice that as the threshold is increased, the criteria for merging regions becomes looser and the amount of regions starts to shrink. | Here we notice that as the threshold is increased, the criteria for merging regions becomes looser and the amount of regions starts to shrink. | ||
| − | The following link provides a precise definition for this algorithm. [http://cobweb.ecn.purdue.edu/~bouman/grad-labs/lab3/pdf/lab.pdf|Image Segmentation Lab] | + | The following link provides a precise definition for this algorithm. [http://cobweb.ecn.purdue.edu/~bouman/grad-labs/lab3/pdf/lab.pdf| Image Segmentation Lab] |
Revision as of 15:17, 18 April 2008
Here is an example of a clustering method used for image segmentation. Here the distance criterion used is the absolute value of the distance between the pixel values. Pixels and their neighbors are chosen from a four point neighborhood and then evaluated for their distances. By adjusting the threshold used to connect pixels, different levels of segmentation are achieved. Here are some results for this algorithm using a simple image.
Original Image:
Image with a low distance threshold:
This strict threshold generated 27,654 connected sets, 36 of which are shown.
Image with a medium distance threshold:
This moderate threshold generated 16,747 connected sets, 41 of which are shown.
Image with a high distance threshold:
This loose threshold generated 11,192 connected sets, of which 23 are shown.
Here we notice that as the threshold is increased, the criteria for merging regions becomes looser and the amount of regions starts to shrink.
The following link provides a precise definition for this algorithm. Image Segmentation Lab
