Imaging through scattering media with machine learning

Abstract

In the areas of biomedical, earth observatory and astronomical imaging, scattering media poses a problem as conventional imaging system are not able to account for light being randomly scattered. Conventional imaging systems would capture a speckle pattern image instead of an undistorted image of the target object. This project proposes a deep learning approach to achieve imaging through scattering media using deep convolutional neural networks. Several tests with different scenarios were conducted to evaluate the viability of such an approach. From the results, it was observed that the chosen deep convolutional neural network architecture exhibited the ability to perform imaging through scattering media.