With the increasing prominence of mobile photography, capturing high-quality images in low-light conditions, especially with flash, remains a significant challenge. This study introduces innovative deep learning techniques to convert flash images into ambient images, with a particular focus on style transfer methods. A novel approach employing CycleGAN for flash-to-ambient image conversion, achieving a mean PSNR of 16.667 on a diverse dataset. Comparative analysis against other models highlights CycleGAN’s superior performance, both in terms of objective metrics and subjective visual quality. This approach showcases promising potential in overcoming the limitations of current techniques, significantly enhancing the realism and quality of images captured in challenging lighting conditions.
