CycleGAN algorithm based enhancement of night vision halation images from multiple sources

Night vision systems are essential for applications such as search and rescue, navigation, and surveillance; however, halation is a common issue with night vision images. This work examines the application of the Cycle-Consistent Generative Adversarial Network (CycleGAN) algorithm to convert various source night vision halation images into their equivalent high-quality, halation-free counterparts. The suggested method utilizes CycleGAN’s cycle-consistency loss to learn the mapping between the halation-affected and halation-free image domains, eliminating the need for paired training data. Even when trained on a variety of datasets, the experiments show how well the CycleGAN model reduces halation artifacts and enhances the overall visual quality of night vision images. The proposed method exhibits a higher peak signal to noise ratio, higher structural similarity index measure, and higher information entropy value, and it is most suitable for the night vision system.