Noise is endemic in microscopy. From the statistical nature of charge carriers to the limitations of detectors, random fluctuations can be an unavoidable addition to signals. Denoisers are therefore a crucial tool for uncovering structures of interest. Now, with deep learning, modern denoisers are more than just smoothing functions. Instead, they can learn about the very structures they are trying to reveal, intelligently separating signal from noise.
This talk will start by explaining the important milestones of deep learning denoisers for microscopy, covering the powerful but costly supervised denoisers and the latest in data-efficient unsupervised denoisers. After that, we will see each of them in action on real volumetric data from a variety of microscopy techniques including light and electron microscopy.
Ben Salmon is a PhD student in the group of Alex Krull at the University of Birmingham, UK. Ben has been developing deep learning-based denoisers for microscopy. His objective is to design denoisers that make minimal assumptions about the nature of the noise and can be trained using only noisy data.
Chad Moore is a multimodal electron microscopist working at Sydney Microscopy and Microanalysis (SMM) at the University of Sydney. Chad is interested in making open-source deep learning image analysis available to users of SMM and has recently been applying these techniques to volume data generated in the facility.
