Innovations in signal/image processing and data analysis in optical microscopy
Modern optical imaging relies on several computational techniques to address different challenges and fundamental limitations, such as noise, limited space-bandwidth product, unwanted color variability and quantification of relevant image features. A wide range of tools spans from classical image processing all the way to advanced deep learning models are now used to enhance the information content in images and the extraction of meaningful and quantifiable parameters. In many cases, machine learning has become the main method of choice for task-specific applications; however, classical image processing techniques still enjoy wide use as general-purpose tools, especially in low-data instances. Due to well-known challenges and limitations to conventional deep learning, researchers now work on emerging techniques, such as explainable AI, physics-informed or physics-supervised learning, known-operator learning, and others that aim to open the black-box of previous models and promise increased interpretability, incorporation of expert knowledge, and faster convergence for smaller datasets.