I don't really follow what your goals are here. Are you really intending to do a forward MNF transform on a spectral library? ENVI's MNF transform tool is meant to be used on spectral images, not non-image spectra (such as spectral libraries). It cannot be used on a spectral library. Although a spectral library does usually contain data from more than one wavelength, all of those wavelength responses are stored in only one "band" of data in terms of the ENVI file format. ENVI's MNF tool cannot understand how to interpret the multiple wavelengths stored in a spectral library file. - P

Thanks for the reply peg. My aim is to collect the different data from near by pixels in a spectraql library which is showing 128 bands and do an MNF transform on those data. your description of "one band" solved my confusion, but i still dont know who to proceed. I need to collect the information of neighbouring pixels and do a MNF forward rotation on it to remove the noise to be precisely soemthing like this Minimum noise fraction transform (MNF) was applied to a spectrally subset reflectance image from the validation set of images taken in 2007. After the MNF rotation the MNF eigenvalue plot was analyzed. The pixel purity index (PPI) function was run on the MNF transform result. The purest pixels were clustered using the n-D visualizer tool with MNF data that used only the purest pixels. The purest pixels are saved as a library of pure endmember collection Whats the input for the MNF? Just an image? Thanks for the information, if you can just explain this because I need to work on my basics. Thanks

Yes, the input to the MNF transform is a spectral image.

thanks i understood till that point but if I do MNF of an image for a given band i will get a plot for Eigenvalu Vs Eigen number, what does it show? Does it rotate entire data of the image for the given band? I need to run the ppi on it, so after forward rotation should go forward? I need to take only those bands who have high eigenvalues so what should i do?(Click OK. ENVI calculates the statistics and the Select Output MNF Bands dialog appears, with each band listed with its corresponding eigenvalue. Also listed is the cumulative percentage of data variancecontained in each MNF band for all bands.) rite? then whichever band has high should be forwarded to PPI Thanks for patience peg

It sounds like you are expecting to appy an MNF transform to one band of a spectral image. That is not how you use the MNF transform. You need to have multiple bands in the input file. The point of the MNF transform is to transform a multi-band image into a different space, in which each output MNF "band" is a combination of the input bands. The new MNF space (as opposed to the original spectral band space) separates the information into some output MNF bands, and the noise into others. So, you can then use only the MNF "bands" that contain information in further processing. It reduces the number of bands you need to include in the processing. It makes no sense to apply an MNF transform to one band. - P

Thanks I got it. Now I have soem other doubt. I did MNF on an image. Then on the result i did PPI both of them did yield me plots. The doubts are. 1. Is the output of MNF an image? while I am opening it, its showing something wierd, lot of chequered flag. 2. Is the result of PPI an image? I know they are because in a journal(which I am referrign showing the MNF as grey image and PPI output as bit image). where i am going wrong 3. To run the n-D visualiser should i use the result of MNF? or the result of PPI because its throwing up some not ROI error. Thanks man