StyleGAN3 algorithm for dehazing nighttime images

Many researchers have worked on improving night vision halation images by using various algorithms, but there is still ample space to use new algorithms to get clear halation-free night vision images. Our research aims to utilize the StyleGAN3 algorithm for night vision halation image generation and also examines both its potential and limitations. The StyleGAN3 algorithm output is analyzed to assess the model’s ability and capacity for creating diverse and realistic night vision simulations. The hazy input image is handed to the image pre-processing process, where the guided filter suppresses haze and retains edges. Contrast-limited adaptive histogram equalization amplifies local contrast and avoids overexposure artifacts. Then, it is handed to the pixel2style2pixel encoder, which maps a noisy halated image into the latent space, and then, it is handed to the StyleGAN3 generator, which rebuilds an enhanced output image with reduced halation. The result shows that it has improved the visibility and image quality compared to existing halation algorithms. Our empirical results show that it has a better information entropy, peak signal to noise ratio, inception score, Fréchet inception distance, and structural similarity index measure than the existing techniques and is most apt for night vision halation images.