Major: Mathematics and Computer Science
Faculty Mentor: Dr. Lynette Boos, Mathematics
We evaluate the ability of self-supervised deep learning for Poisson denoising of Single-Molecule Localization Microscopy (SMLM) in addition to the impact denoising can have on the ability to locate molecules within the Single-Molecule Localization Microscopy images. SMLM images are predominantly corrupted with Poisson noise. There is a need for a superior technique to provide accurate SMLM images in order for scientists to gain a better understanding of the functions of live cells at the nanoscale. By denoising SMLM images prior to the images undergoing the current state- of-the-art super-resolution techniques, we create a less corrupted version of SMLM images. As a result, the exact locations of the molecules in the images can be determined with more accuracy and precision. We denoise SMLM images utilizing only the original noisy images as training data with a Self-Supervised Deep Learning model. By modifying the previous Self-Supervised techniques that have been successful in denoising images with Gaussian noise, we remove Poisson noise from SMLM images.
4-22-2020 12:00 AM