Cloud removal from aerial images using generative adversarial network with simple image enhancement

• Main Authors: Pacot, Mark Phil B., Marcos, Nelson
• Format: text
• Published: Animo Repository 2020
• Subjects: Imaging systems—Image quality; Drone aircraft in remote sensing; Remote sensing—Atmospheric effects; Computer Sciences; Software Engineering
• Online Access: https://animorepository.dlsu.edu.ph/faculty_research/2739
• Institution: De La Salle University

The atmospheric condition of the presence of clouds is one of the biggest problems in most aerial imaging systems. It degrades the visual quality of images leading to the loss of information for ground scenes. Hence, an effective cloud removal algorithm is a significant factor for this kind of problem and other related applications. The proposed cloud removal technique using the generative adversarial network with simple image enhancement (SIE-GAN) is a useful tool in removing cloud formations, most notably in images acquired using Unmanned Aerial Vehicle System (UAVs). This technique showed flexibility in performing the given task with satisfactory results, which is a gauge based on No-Reference Image Quality Metric, specifically the Perception-based Image Quality Evaluator (PIQE). Also, the proposed algorithm outperformed some of existing cloud removal algorithms by producing a better quality output when tested on the too-cloudy satellite images. Overall, the authors introduced a new frontier in generating cloud-free aerial images and added a valuable contribution to the array of cloud removal algorithms. © 2020 ACM.