Hi folks, I'm using Envi to run a series of feature extractions on very high resolution mosaic aerial photos (5cm and under resolution), and I'm running in to some problems. I'm specifically trying to isolate two classes from my image - vegetation from bare ground, but my imagery is only visible spectrum RGB, so I don't have very many bands to play with. Also, since I'm not isolating unnatural from natural features, there's not much by ways of purely spatial information (feature size, shape, etc.) that is unique between the two classes. Basically I'm left with texture and spectral information that can differentiate the two classes, but I'm being confounded by the fact that as this imagery is so high-rez, the shadows being cast by the vegetation should almost be considered their own third class, but there doesn't seem to be anything spectrally unique about much of the 'shadow' objects - in some parts of the image, the bare sand is as dark or textured as the shadows in another part of the image, so no one rule captures all the shadows without also capturing too much sand. I've broken up the image in to 10 separate areas to try and isolate for the the variation in focus, exposure, and general lighting present in this mosaic, but even within the smaller regions I'm still having a hard time coming up with ways to help the software accurately differentiate between the vegetation and the sand accurately, even though I can see with my eyes which is which. Is there some pre-processing step for RGB imagery that I haven't thought of? I've tried every stretch that makes sense. Is there something that I could write in IDL that would make the process more accurate?