Image Classification Tool ArcGIS 10

[GeospatialTechnology]

Hello,

I really want to learn to use the Image Classification Tool released with ArcGIS 10.

Does anyone know of any resources that will help me to fully understand this tool?

Are there any online tutorials?

Thanks for any help!

[MohammadDashti]

Mark,

Check this web seminar, I hope it will be useful for you.
http://training.esri.com/gateway/index.cfm?fa=catalog.webCourseDetail&courseid=1949

Regards
Mohammad

[GeospatialTechnology]

Hello,

Is there anymore information on the image classification tool?

The web seminar was an excellent introduction but I need more information!

Im selecting an area of land that is dark in colour. However when I run a classification I end up with shadows from trees, buildings, gullies, etc being classified as this land type. Im not sure if theres anyway around this?

Im using CIR aerial photographs.

[JeffreySwain]

I would consider creating training samples of the shadows and try to collect them as a separate feature.  If you are still having issues distinguishing between the two, perhaps clipping those portions out of the raster and then processing them independently to determine the difference and merging the results. 

There are other blogs discussing the image classification toolbar or this one about land cover.

[GeospatialTechnology]

Thanks Jeff, yes ive read those blogs already, but thanks for suggesting them.

What I guess Im trying to do is understand whats going on in the background of image classification so i can understand the process better.

For example Im looking at CIR aerial photographs.

In image classification is it looking at the different RGB values for each of the bands and then return pixels of similar values?

[JeffreySwain]

The image classification tools are looking at ranges of the pixels selected as the training, not necessarily the combination of one pixel next to another.  That would be object oriented classification, where the arrangement and the pixel values are considered. What is probably happening with your data is that based on the training samples created the shadows might fall into another class like rooftops or roads creating the confusion.  Since the algorithm is only looking at the pixel values, such an issue can and does occur.  The only way to improve the classification is to improve the quantity and quality of the training samples or masking out some of the raster in order to analyze again. Using the image classification toolbar and then looking at the histograms created by your training samples can help you understand what the computer is looking at when it classifies the pixels.  I hope this helps explain what is happening, it is an over simplified explanation, but the concept is that pixel values in band combination are what it is looking at.  There are many journals that discuss the difference between pixel based and object oriented, this is one.

[JakeSimpson]

I am using ArcMap 10.1 with Spatial analyst extension, which is for the most part really excellent. 

The issue is in using 'Image classification tool', specifically when trying to analyse the training samples using scatterplots/histograms/statistics from the 'Training sample manager'. When I try to analyse the sample, I get an error message saying 'out of memory'.

This does not seem to be resolved by freeing space in the geodatabases or by attempting the operation in a new/empty session/geodatabase. I also tried using smaller training samples, but I still get the same error. I have also tried increasing the amount of available memory, but this was also ineffective.

Any ideas/help would greatly appreciated as this is a persistant and very irritating problem.

Thanks

Jake