Hi folks, I'm trying to run a feature extraction on a large aerial mosaic using the custom colour space bands (HSI) for the segmentation step. However, I'm experiencing this very strange artifacting that I'm having trouble sourcing. I'm having a hard time trying to explain it, so hopefully the screenshot I've attached of a small portion of my image showing the segmentation preview window will help clarify. I'm relatively green on how this forum works, so forgive me if I don't embed the image properly. If the link doesn't display the image in text, there's a direct link to the screenshot in my photobucket account account below it. [URL=http://s1100.photobucket.com/user/asylumdown/media/file_zpsa0567bf3.jpg.html] [/URL] http://i1100.photobucket.com/albums/g411/asylumdown/file_zpsa0567bf3.jpg It's almost as if the image has been cut in to strips, and drastically different HSI values have been calculated for each strip, creating this strange banding in the segmentation result. It's making it seem as though completely different scale and merge values are being applied in perfectly linear bands within the same image. Obviously, this is making it impossible to segment my image, as an appropriate scale and merge level for the features I'm looking for in one band makes the adjacent strip either over or under segmented. The image is a basic RGB colour composite that was obtained by an automated unmanned aerial vehicle, and is a mosaic of 140 separate images. The image was produced by a third party and I'm not sure of the software they used to create the mosaic, so I thought at first it might have been an artifact from that process, but I've gone back and looked at the flight lines and image way points, and while the flight lines did run east/west (or horizontal relative to the image) these bands seem to have no relationship to them or where the original images overlapped. For example, the bands are perfectly straight, and the flight lines were not, and there are bands in the extreme north and south ends of the image where there was no north/south overlap in the original imagery. This problem disappears when the segmentation is done based on the RGB bands. Other details: Texture Kernal size of 3 (default), edge segment algorithm, full lambda schedule merge algorithm. The attached screenshot was at a scale and merge level of 50. My main question - What is the source of this problem? Is there something wrong with the image from the way it was processed in to a composite, or is this something Envi is doing? My second Question - is there anything I can do to fix it? Thanks in advance