This talk presents a method of denoising Cryo-EM images without class averages. Collections of images are viewed as a sections of the trivial vector bundle L^2(R^2) x SO(3). Representation theoretic considerations identify a low dimensional subspace containing the signal of this section, the images are denoised by projecting onto this subspace. Under assumptions the method scales snr linearly with the number of images and recovers true representations of the images.
Denoising by Global Projection
Jim Delfeld - University of Texas
Apr 30 2014 - 4:00pm
110 Fine Hall