Dear all,   I have performed a maximum likelihood classification on an ASTER image using ground truth ROI’s. What I would like to do is validate the classified image, by using 95 % of my ROI’s for the classification and 5 % for the validation. I have randomly selected 95 % of my ROI’s (by generating a random sample of my ROI’s), but the problem is I need the other 5 % for my validation. I made an image of the 95 % ROI’s and tried to substract them to get the 5 %, so that I can use that resulting image as a ground truth image for the confusion matrix (and just ignore unclassified pixels), but this doesn’t work (I tried with masking and with band math). Who has any advice? (I also tried the other way around, selecting 5 % for the validation image - but then getting the 95 % of the ROI's for the classification becomes a problem) Perhaps there is an easier way to do this validation? Because actually I would like to run several classifications with varying 95 % ROI’s (using 5 % for validation), so that I get a good idea of the accuracy of the classification.   Thank you.