RestoreNet: a deep learning framework for image restoration in optical multi-aperture imaging system

Imaging blur is an inevitable problem because of the low response to medium frequency in optical multi-aperture imaging system, in which Wiener filtering is usually used in implementing image restoration to obtain clear high-resolution images. Recent notable developments in the field of deep learning have opened up exciting avenues inspiring us to use data-driven approach for image deblurring in optical multi-aperture imaging system. In this paper, a deep learning framework named RestoreNet, which is based on a U-shaped convolution neural network, is proposed to replace the general Wiener filtering for image restoration. Numerical simulation and experiment results show that RestoreNet could recover the imaging map from system successfully, just like Wiener filtering does. However, RestoreNet only needs one dataset containing a few images for training, and shows strong image restoration ability without the point spread function or optical transfer function of system in testing, as well as the priori information of object and noise. As a result, RestoreNet is an effective alternative in image restoration of the optical multi-aperture imaging system.