Imaging through dynamic thick scattering media based on deep learning

Abstract

When light passes through a dynamic scattering media, traditional scattering media imaging methods cannot effectively achieve image recovery due to the change of the dynamic scattering medium with time and the limitation of memory effect. In the paper, a neural network model was constructed to study the scattering image recovery effect of dynamic scattering media with different concentrations, and the differences of image recovery between this neural network model and standard convolution and depth wise separable convolution are compared and analyzed by two indexes, Peak Signal to Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM). The results show that this neural network outperforms the other two neural network models in the reconstruction of scattered images.