Defocused Image Restoration as an Inverse Problem by Network Inversion

摘要

Recently, the necessity to solve an inverse problem has increased in the various fields. The solution for the inverse problem by the neural network, which is called the network inversion technique, was proposed. On the other hand, the image restoration problem for recovering the degraded image to the original one is so important that a number of methods have been proposed. The image restoration can be defined as a kind of inverse problem of the image degradation process. Therefore, we introduce the network inversion method to solve the image restoration as an inverse problem. In this study, we perform the simulation of the image restoration by the network inversion. To compare the effect quantitatively, we evaluate the restored image by the ratio of the spatial-frequencies.