High fidelity image reconstruction of light passing through scattering medium based on convolutional neural network

Abstract

Optical imaging through scattering media such as ground glass, fog, biological tissues, etc. has always been a widely used and challenging task in the optical field. Compared with traditional imaging methods such as transmission matrix and optical phase conjugation, deep learning has shown great potential in this field because of its simple device and fast reconstruction speed. In this article, we developed an algorithm based on convolutional neural network to realize imaging through scattering media and applied this algorithm to recover complex images. The speckle images of the original images are obtained through a speckle generation program, and then the speckle images and the original images are input into the neural network in pairs for training. Finally, the reconstructed speckle images can be obtained by using the trained neural network. In the numerical simulation, we proposed two indicators, peak signal-to-noise ratio (PSNR) and structural similarity (SSIM), to evaluate the quality of reconstructed images. The results show that our method can restore highfidelity images. This new image reconstruction method provides new ideas for research in the fields of astronomy and biomedicine.