The paper addresses the challenge of image dehazing, crucial for improving visibility in computer vision applications. Traditional methods such as Dark Channel Prior (DCP) and Color Attenuation Prior (CAP) rely on handcrafted features, but struggle with dense haze and diverse conditions. Deep learning, specifically Convolutional Neural Networks (CNNs), has enhanced dehazing, with models such as DehazeNet and AOD-Net showing significant improvements. The paper introduces a novel SwinDehazing-based U-Net architecture, integrating Swin Transformer and U-Net for robust multi-scale feature extraction and spatial resolution recovery. Key contributions include performance on benchmark datasets, a detailed analysis of the model’s components, and enhancements such as replacing GeLU with ReLU and employing shifted window partitioning. The proposed SwinDehazing model demonstrates higher dehazing performance, effectively handling varying haze densities and complex scenes.