Displacement-sensible imaging through unknown scattering media via physics-aware learning

Imaging through unstable scattering scenes is a challenging task and the focus of attention. More computational techniques and optimization methods have been introduced in objects recovery hidden behind opaque scattering media. A combination of the optical memory effect (OME) prior and deep learning (DL) has been demonstrated in scalable imaging through unknown diffusers with high fidelity. Here, a displacement-sensible imaging method is proposed to reconstruct the hidden objects with unseen depth-of-field (DOF) positions, and the reconstructed objects with different displacements are aligned with the practical imaging size. Related physics priors to DOF are excavated by utilizing the speckle-correlation magnification and developing a flexible DL framework to recover the objects with variable DOF positions. This flexible physics-aware learning approach gives a heuristic to complex imaging problems with specific application scenes.

Automated summary

This article proposes a displacement-sensitive imaging method, which combines the optical memory effect prior and deep learning to recover hidden objects behind opaque scattering media. By utilizing speckle-correlation magnification to excavate physics priors related to depth of field, and developing a flexible deep learning framework to recover objects with different depth of field positions.