Principal Component Analysis (PCA) can drastically denoise STEM spectrum-images but might distort or cut off the important variations in data. The present paper analyzes various approaches to estimate such deviations and compares them with the simulated data. A spiked covariance model by Nadler (2008) appears to be most appropriated for application in STEM spectrum-imaging.
Copyright © 2017 Elsevier B.V. All rights reserved.