[UW-GIS-L] eCognition vs. Feature Analyst

[cwayne at u.washington.edu]

Thanks Stefan.  This is very helpful.  My real question is, what is the most effecient us of our budget: taking time to learn it myself or contracting it out to some experienced folks at USGS?  Given what I have heard, I suspect the latter is my best option.  Thanks again!


Chris Wayne
GIS Instructor
University of Washington 
Educational Outreach
cwayne at u.washington.edu
http://faculty.washington.edu/cwayne

On Mon, 24 Jul 2006, Stefan E. Coe wrote:

> There are some key differences between the two. In simple terms, Feature 
> Analyst is a black box classifier. Its easy to learn, easy to use, and can 
> extract features pretty well using high resolution imagery. It works very much 
> like a traditional supervised classification: you digitize areas of interest 
> that represent the features you wish to extract and these sites are used to 
> parametrize the classifier. The software has some cool features that helps you 
> iteratively train the classifier, e.g. you can select falsely classified 
> features (both errors of omission and commission) to help train the classifier. 
> All in all, it works reasonably well with one caveat: the results are very 
> dependent on the granularity of your classification and the heterogeneity of 
> your scene. I think feature analyst would work quite well if you were trying to 
> map general vegetation classes such forest and grass but maybe not so great if 
> you wanted to break it down into species or seral stage.
>
> As Luke says, eCogntion is very complex, powerful and a bit
> overwhelming. The actual classification process is much more up to the user and
> there are many different directions one could go (hence the overwhelming
> factor). Its primary purpose is to segment the image into objects (polygons)
> which form the basis of the features (as opposed to pixels) to be classified. 
> These object can than be used in a traditional supervised classification 
> (called the standard nearest neighbor classifier) or in a rules based 
> classification (e.g. label this object grass if the mean value of band 1 is 
> less than 50) or a combination of the two. You can also do a hierarchical 
> object classification where you have segmentations of different levels of 
> heterogeneity and hence larger super objects and smaller sub objects that are 
> spatially nested. These can then be used to hierarchically break down your 
> image into different classification levels. For example, your top level (larger 
> objects) could be used to separate urban and vegetation objects. Since the 
> sub-objects are nested within the super objects, you can create parent/child 
> rules so you can apply classification rules to your sub-objects based upon 
> their super-object membership.
>
> Wow, I did not intend to write that much. I hope at least part of it makes 
> sense. The bottom line is that if you think you can get away with feature 
> analyst, I would lean in that direction because eCognition takes a considerable 
> time investment to become really proficient. I would be happy to answer any 
> other questions. Good luck,
>
> -Stefan
>
>
> Stefan Coe
> GIS/Remote Sensing Analyst
> Urban Ecology Research Lab
> University of Washington
>
>
> Stefan Coe
> GIS/Remote Sensing Analyst
> Urban Ecology Research Lab
> University of Washington
> 206-616-9379
>
> On Mon, 24 Jul 2006, Luke Rogers wrote:
>
>> eCognition is a big, ugly, complicated, unpredictable and not well
>> documented yet very cool and powerful piece of software. I shouldn't but
>> I'll put Stefan on the spot since he has used both... I have only used
>> eCognition. I put the eCognition user guide here
>> (http://www.ruraltech.org/downloads/UserGuide.pdf) which is all the
>> documentation you get.
>> 
>> -Luke
>> 
>> -----Original Message-----
>> From: uw-gis-l-bounces at mailman1.u.washington.edu
>> [mailto:uw-gis-l-bounces at mailman1.u.washington.edu] On Behalf Of
>> cwayne at u.washington.edu
>> Sent: Monday, July 24, 2006 2:02 PM
>> To: uw-gis-l at u.washington.edu
>> Subject: [UW-GIS-L] eCognition vs. Feature Analyst
>> 
>> We are considering which software to use for an alliance-level veg mapping
>> project, based on color 1-m orthos.  One camp is pushing for eCognition,
>> while another favors Feature Analyst.  My reservations for both are based on
>> accuracy and the learning curve behind each.
>> 
>> My general impression is that eCognition is more accurate, but harder to
>> learn and more expensive.  Can anyone share their experiences, or point to
>> me to case studies using either of the above?  Thanks!
>> 
>> caw
>> 
>> Chris Wayne
>> GIS Instructor
>> University of Washington
>> Educational Outreach
>> cwayne at u.washington.edu
>> http://faculty.washington.edu/cwayne
>> 
>> 
>> 
>> _______________________________________________
>> Uw-gis-l mailing list
>> Uw-gis-l at u.washington.edu
>> http://mailman1.u.washington.edu/mailman/listinfo/uw-gis-l
>> 
>> _______________________________________________
>> Uw-gis-l mailing list
>> Uw-gis-l at u.washington.edu
>> http://mailman1.u.washington.edu/mailman/listinfo/uw-gis-l
>> 
>
>
>
>
>
> _______________________________________________
> Uw-gis-l mailing list
> Uw-gis-l at u.washington.edu
> http://mailman1.u.washington.edu/mailman/listinfo/uw-gis-l
>