Re:[erdas-l] supervised classification in Imagine versus ArcGIS

[George Mueller-Warrant]

I'm now up to the next thing I can't figure out. Using Wan's suggestion for a series of single-class rasters, I get signature files that generate maximum-likelihood classification rasters very similar to those from ArcGIS using the "equal probability" option in ArcGIS. What I can't do is to get ERDAS Imagine to match results for the "sample frequency" weighting in ArcGIS. I've tried both checking and unchecking the "Use signature probabilities (a priori)" option in Supervised Classification in Imagine, but get the same results in both cases. I also got the same results when I used "Mahalanobis Distance" rather than "Maximum Likelihood" for the Parametric rule. Any ideas on what step(s) I am missing?

Thanks
---
You are currently subscribed to erdas-l as: erdas-l-gar...@googlegroups.com.
To unsubscribe send a blank email to leave-31091599-120484586.1d5...@lyris.dundee.net

[George Mueller-Warrant]

I solved the "sample weighting" problem by going into the signature editor and manually setting the entries in the probability column from 1.0 down to numbers proportional to class frequency - something that ArcGIS does automatically for you. And now results from ArcGIS and ERDAS Imagine match in about 99% of cells. I presume the small discrepancies (~1.4% in the only case I've finished so far) have to do with differences in exactly how calculations are run behind the scenes.

---
You are currently subscribed to erdas-l as: erdas-l-gar...@googlegroups.com.

To unsubscribe send a blank email to leave-31092378-120484586.1d5...@lyris.dundee.net