Imaging through scattering media via support vector regression.

Abstract A clear image of observed object may deteriorate into an unrecognizable speckle pattern when encountering with heterogeneous scattering media, thus it is necessary to recover the object image from the speckle pattern. Here, a machine-learning-based support vector regression (SVR) method for imaging through scattering media is experimentally demonstrated. The proposed method learns inverse scattering function (ISF) with known object-and-speckle pairs, then reconstructs unknown object with the learned ISF. Essential normalization preprocessing is pre-performed before learning the ISF. Experiments show that more training pairs lead to more accurate ISF and higher reconstruction fidelity. The proposed method provides a general solution for imaging through scattering media and is expected to has its potential applications on inverse problems, such as phase retrieval. Highlights • A machine-learning based support vector regression method for imaging through scattering media is proposed. • The method is experimentally validated to be data-driven and feature-driven, as well as noise robust to some extent. • The method realizes inverse scattering imaging, and has its potential in inverse problems such as phase retrieval. [ABSTRACT FROM AUTHOR]