Motion-informed wide-field dynamic imaging through scattering layers

Exploiting multiple light scattering holds great potential for deep optical imaging. However, practical implementations often face trade-offs in imaging speed, field of view (FOV), system complexity, or impose special requirements on the sample. We propose localization and stream fusion (LSFusion) for wide-field imaging of moving objects through scattering media with a simple experimental setup. LSFusion treats the motion of hidden objects, which induces random activation of the point sources constituting the object across the FOV, as an equivalent modulation required in the nonnegative matrix factorization (NMF) technique. This enables large-FOV imaging using scattered light at the camera’s native frame rate without the need for modulation devices. Furthermore, by jointly reconstructing the video sequence, LS-Fusion recovers widely separated point sources beyond the limits of conventional NMF. Experimental results through a 200-µm-thick mouse brain slice demonstrate a FOV 5.7 times larger than the memory effect range (MER), enabling high-quality dynamic imaging at 50 frames per second (fps), even for hidden objects with point sources spaced far beyond the MER.