The MF floating-point results help estimate the relative degree of match to the reference spectrum and the approximate sub-pixel abundance, where 1.0 is a perfect match. So MF is abundance abundance per endmember and infeasibility can be used to estimate the accuracy (or how feasible) the MF value is. You can always run LSU with your endmembers to get those results.
For classifying MTMF results, you can use the 2D scatterplot to view an MTMF band vs its infeasibility score to select pixels in the scatterplot. Export these to ROIs and then you can create a classification image from the ROIs.

Hello everyone,

I am trying to implement the same procedure; to find out the pixels with high MF and low infeasibility and then to validate those data. The problem with using a 2D scatterplot is the overlay of the rasters... Furthermore, about the above-mentioned response; to use LSU, do you suggest to use MTMF results at the LSU algorithm?Thank you very much in advance for your help!

Sorry, no that is not what I meant and it was not clear.
MTMF and LSU *both* provide sub-pixel abundance per pixel. The difference is that MTMF is designed to find one material abundance against 'background' of all other pixels/materials. So it must look at each endmember separately. LSU is designed to be used with *all* relevant endmembers and find abundance per pixel of all endmembers at once. However, this only works well if all endmembers are well defined. If you don't know about some endmembers, then you should use MTMF. Also, LSU doesn't create the infeasibility measure.
I am not sure what you mean by the problem with using a 2D scatterplot. Could you explain?

Thank you very much for your response. Now I see what you mean about the two algorithms. Actually, I believe that MTMF is the appropriate method for my case. The problem is that I can't find a way to represent the results and validate them. I thought about creating a 2D scatterplot as you suggested and correlate infeasibility - MF score. Then I chose the appropriate pixels (low infeasibility and high MF score), and I create classes of ROIS. However, the pixels that I choose are overlapped (for example montmorillonite with kaolinite). So the raster that I create from all those classes is a "product" of how I will overlap the pixels. I don't know if you see what I mean..