Revolutionary Dehazing Advances: A Comparative Study

Haze is a challenging phenomenon in image analysis, severely obstructing the accurate identification of objects and deteriorating vital features in images. The obscured object information, embedded in the three-dimensional matrix, becomes indistinct due to the presence of haze, resulting in reduced contrast between foreground and background elements. To address this issue, numerous mathematical and deep learning techniques have been proposed, aiming to extract meaningful information and eliminate noise induced by haze. This paper presents a concise representation of the latest dehazing techniques utilized in image analysis. The primary cause of haze is the distortion of atmospheric (or source) light caused by tiny particles and diffused materials in the environment. In this review, we survey various state-of-the-art approaches used to combat haze-related challenges. With the focus on generative adversarial network (GAN) networks the mathematical methods are explored, leveraging advanced algorithms to filter out valid information from hazy images. Additionally, the growing prominence of deep learning techniques in dehazing is discussed, highlighting their ability to learn complex features and effectively remove haze-induced noise.