Fluorescence imaging through scattering media with robust matrix factorization

The applications of fluorescence microscopy are greatly limited when the target is obscured by scattering in biological tissues. Extracting meaningful features from random speckle patterns is crucial for achieving high-quality imaging through scattering media. In this work, we present RNP, a deterministic framework that utilizes sparse representation to extract object features from scattering speckles amid strong background disturbances, enabling high-quality reconstruction under challenging conditions. This framework was demonstrated on a standard epi-fluorescence microscopy platform using various samples, achieving a large reconstruction field of view, an extended depth of field, and enhanced image clarity. We anticipate that RNP will emerge as a promising tool for advancing non-invasive scattering imaging in biological research.