Summary: Denoising in Fluorescence Microscopy Using Compressed Sensing With Multiple
Reconstructions and Non-Local Merging
Marcio Marim, Elsa Angelini and Jean-Christophe Olivo-Marin
Abstract-- The cross-dependency of noise level and photobleaching
in microscopy was discussed in a previous work and an efficient
compressed sensing (CS) method was proposed to simultaneously
reduce the noise level and the photobleaching. Here we present an
improved CS denoising framework for fluorescence microscopy images,
exploiting Non-Local means filtering to merge multiple reconstructions.
This framework enables high-quality reconstruction of low exposed
microscopy images based on random Fourier sampling schemes and
multiple CS reconstructions. Practical experiments on fluorescence
images demonstrate that even performing 10% of the measurements,
the signal-to-noise ratio can be significantly improved while keeping
reduced exposure time, preserving edges and the image sharpness.
Index Terms-- Compressed sensing, denoising, fluorescence mi-
croscopy, NL-means filtering.
In biological fluorescence microscopy, cellular components of
interest in the specimen such as proteins are typically labeled with a