Exploring the range of optical memory effect by deep learning

Among the various methods for imaging through scattering media, the speckle autocorrelation imaging method based on optical memory effect (OME) has gained significant attention for its non-invasive single-shot imaging capability through scatter media. However, the imaging range of this method is limited by OME. This paper presents a physics-informed deep learning strategy that establishes relationships among different linearly-shift-invariant subsystems based on OME to explore the range of OME range. By leveraging the feature extraction capabilities of deep learning, the proposed approach recovers sidelobes from the speckle autocorrelation patterns of objects in several OME regions. Then, a phase retrieval algorithm is employed to achieve object reconstruction. In the future, the approach can be extended to integrate different small regions of the object plane into a big one which can be linear shift invariant.