Hi, I have Landsat TM image and I am trying to perform linear spectral unmixing on it. The image is about 9 degrees oriented and therefore a dark background is also shown in the image display. When i try unmixing the image, I encounter a number of problems. First, MNF transform curve indicates image dimensionality as only two. Secondly, PPI detects a lot of regularly spaced pure pixels in the dark background area. Third, the n-D visualizer does not show any clusters at all, just very few (two or three) points. Whereas if I take a rectangular subset of the image, and re-run the process, these problems are solved i.e. MNF curve shows higher dimensionality, and good clusters of points in n-D visualizer. So apparently the problems are caused by the dark area in the display due to image orientation, but I could not figure out exactly how the dark area pixels cause problems in PCAs and hence in the MNF transform. Can some one please give me clues? Help is ugently needed. A lot of thanks in advance.