RETRACTED ARTICLE: Learning-based approach to underwater image dehazing using CycleGAN

In this paper, we have proposed a novel deep learning-based underwater image dehazing method. The images taken under water are affected by light transportation properties like refraction, reflection, scattering, and absorption. Deep learning techniques give great results in the removal of haze in underwater degraded images. But the major challenge we are facing is the need of many images from the same location to train the model. The proposed system addressed this challenge by using CycleGAN, capable of producing synthetic underwater images from RGB in-air images, and is validated using images taken from different locations and with different characteristics. Qualitative evaluation has been done based upon visual analysis. Quantitative analysis is based upon the calculation of various parameters such as MSE, RMSE, and Euclidean distance. In addition, SSIM Loss has also been evaluated and compared. The model is tested for underwater images taken under various circumstances, and the quantitative and qualitative results obtained for these images are quite satisfactory.