🚀 go-pugleaf

As promised here is a summary of the replies I got to from my query
concerning the segmentation of remotely sensed images. The original
posting was sent to: sci.image.processing, comp.infosystems.gis and
IMARS-L.

====================================================================

The evaluation and comparison of segmentation techniques has long
since been a problem:

In article <1992Aug5.185035.21313@Arco.COM> opus@arco.com (Dr. Steven 'Wild Man'Wong)
> The problem with evaluating standard segmentation processes is that we
> have too much of the 'See ma, no hands' syndrome that also affects the KBS
> (knowledge based systems)  community.  Unfortunately, I also was part of
> that crowd.  What we need is a public accessible place with standard
> benchmarks and algorihms that everyone can experiment  with, test, and
> compare against.  This idea has been proposed again and again but still
> hasn't gathered any steam.  The standard picture of 'the girl' just
> doesn't cut it.

Peter Sincak mailing to IMARS-L argued that it is difficult to establish
definitions of edges, homogenous areas, clusters ... and it is therefore
difficult to evaluate segmentations.

I think part of the problem in complex scenes, is that you can have
more than one correct segmentation --- it really depends on what your user 
wants from a decomposition of the image.

=====================================================================

Eric Majani recommends the following reference:

Hierarchical Segmentation using a composite criterion for remotely sensed
imagery
Goldberg-M and Zhang-JY
Photogrammetria Vol 42 issue 3, pp. 87-96, 1987

=====================================================================

Janet Franklin, in reply to my comp.infosystems.gis posting suggested that 
the work done by Tailor, Cross, Mason, Hogg et al on the knowledge base 
interpretation of remotely sensed images could be useful.

Their work is based on a map guided segmentation using high level domain
specific knowledge to improve the acuracy of the image decompostion.

I have a complete set of reference for that work ... email me if you are
interested.

=====================================================================

Thanks to all those who replied to my postings ....

Julien Flack,
University of Leeds.
email: julien@scs.leeds.ac.uk