Non-invasive imaging through strongly scattering media based on speckle pattern estimation and deconvolution

Abstract

Abstract Imaging through scattering media is still a formidable challenge with widespread applications ranging from biomedical imaging to remote sensing. Recent research progresses provide several feasible solutions, which are hampered by limited complexity of targets, invasiveness of data collection process and lack of robustness for reconstruction. In this paper, we show that the complex to-be-observed targets can be non-invasively reconstructed with fine details. Training targets, which can be directly reconstructed by speckle correlation and phase retrieval, are utilized as the input of the proposed speckle pattern estimation model, in which speckle modeling and constrained least square optimization are applied to estimate the distribution of the speckle pattern. Reconstructions for to-be-observed targets are realized by deconvoluting the estimated speckle pattern from the acquired integrated intensity matrices (IIMs). The qualities of reconstructed results are ensured by the stable statistical property and memory effect of laser speckle patterns. Experimental results show that the proposed method can reconstruct complex targets in high quality and the reconstruction performance is robust even much less data are acquired.