1

After a supervised classification in ERDAS IMAGINE I have some undefined "ghost classes" in the attribute table. Also have a look at this youtube video at about 7:12.

I'm wondering why they arise and how they affect the classification result and the accuracy assessment?

My first consideration was that they arise within the merging in the signature Editor. Or is it to pixels, that could not be categorized during the classification process because the possibility of a membership to one ore another class is not high enough?For instance I used the maximum likelihood classifier which categorises the pixels depending on the possibility of being a member of a specific class. In the latter case should they be marked in some way? They would also have to be taken into account in the accuracy assessment because the pixels are still there but were not assigned to a class.

| improve this question | |
1

As best I can tell it would make no difference aside from annoyance as the table indicates 0 pixels are classified with these "ghost values". Hence the term. It appears that the pixel values are continuous so if you leave a gap --say a class for 3 and a class for 5-- then it will fill the gap with zeros in the table. In this example class 4.

| improve this answer | |
  • 1
    So do you think the problem lies within the signature file and the merging of the spectral classes (within the AOI) to one "main class"? How can I correct these classes, to get rid of them? – parallax Jan 8 '15 at 14:08
  • They don't actually exist as no pixels have the value. They are not in the data so nothing to get rid of. They are merely in the table. – If you do not know- just GIS Jan 8 '15 at 17:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.