Restoration Algorithm of Blurred UAV Aerial Image Based on Generative Adversarial Network

Abstract

In the process of using visual devices to perform tasks, the image information acquired by unmanned aerial vehicle (UAV) often becomes blurred due to the jitter of its own camera or the movement of objects in the scene, which seriously affects the use of images and subsequent processing. To solve this problem, the paper proposed a single blurred image restoration in an end-to-end method. This method is based on generative adversarial network (GAN), using the improved scale-recurrent network (SRN) as generator and the critic in DeblurGAN as discriminator. Entire network structure is relatively simple, and the trained model has strong generalization ability. Results from numerical experiments show that compared with the Deep-Deblur, SRN and DeblurGAN, the proposed method improves the recovery efficiency of blurred image.