Large-scale Simulated Dataset for Aerial Image Deblurring based on DOTA Dataset
Yuanfa Zhang, Suhe Wang, Weili Li, Xiaoqing Yin, Jinsheng Deng · 2020
Due to the vibration of the aerial imaging platform or the interference of atmospheric transmission, aerial images may be blurred and the image quality may be affected. However, the current traditional image deblurring algorithms are limited in processing capabilities and difficult to solve complex blur problems. Although deep learning methods provide inspirations for aerial image deblurring, large-scale dataset is needed to facilitate the training of deep networks. In this paper, a deblurring dataset generation method for aerial images is designed based on the DOTA dataset, which considers multiple degrading factors and efficiently generates a large number of simulated blurred images. The proposed method can handle the lack of training data and help to recover blur aerial images.