GAN-based Color Correction for Underwater Object Detection
Takafumi Katayama, Tian Song, Takashi Shimamoto, Xiantao Jiang · 2019
Underwater images suffer degradation from light propagates through water. The degradation by light propagation not only affects the quality of underwater images but limits the ability of object detection. In this paper, we propose a color correction method by cycle-consistent adversarial networks (CycleGAN) for an improvement of underwater object detection. The proposed CycleGAN is implemented with a polynomial loss function including adversarial loss, cycle-consistency loss, and structural similarity index measure (SSIM) loss. Thereby, underwater images are generated as output images including the content and structure of the input images. Simulation results show that our method achieves an improved 7% mean average precision in underwater object detection, compared with a conventional method.