An Image Restoration Method for the Sparse Aperture Optical System Based on Deep Learning

In the sparse aperture optical system, the decrease of modulation transfer function (MTF) in the mid-to-high space frequency region inevitably leads to the image blur. To obtain high-resolution images, the Wiener filtering and blind deconvolution algorithms are used to restore the imaging results. This paper proposes a deep learning image restoration method based on a multi-layer progressive image restoration network (MPR-Net) for the sparse aperture system. A dataset was constructed by simulating imaging to train the network, and the image restoration effect of MPR-Net was compared with that of Wiener filtering. Numerical simulation results show that MPRNet demonstrates strong image restoration ability and good generalization ability with less dependence on the dataset.