Power Iteration Denoising

Abstract

We present a simple method for image denoising called power iteration denoising (PID). PID finds a low dimensional embedding of the image data using a truncated power iteration on a normalized pair-wise similarity matrix generated from the image. This embedding turns out to be an effective denoising algorithm outperforming the widely used non-local means algorithm. We apply this method to the denoising of noisy digital camera images producing visually pleasing results.