DehazeNet is an end-to-end deep learning system designed for single image haze removal. The key challenge in this task is estimating the medium transmission map of a hazy image, which is then used to recover a haze-free image through atmospheric scattering models. DehazeNet employs a Convolutional Neural Network (CNN) architecture with specialized layers to embody established assumptions and priors in image dehazing. Specifically, Maxout units are used for feature extraction, and a novel nonlinear activation function called Bilateral Rectified Linear Unit (BReLU) is proposed to improve the quality of the recovered image. The paper establishes connections between DehazeNet components and existing methods, demonstrating superior performance over state-of-the-art techniques while maintaining efficiency and ease of use. Experiments on benchmark images show that DehazeNet outperforms other methods in terms of both quantitative metrics (e.g., MSE, SSIM, PSNR) and qualitative evaluations, particularly in handling challenging scenarios such as sky regions and noise robustness.DehazeNet is an end-to-end deep learning system designed for single image haze removal. The key challenge in this task is estimating the medium transmission map of a hazy image, which is then used to recover a haze-free image through atmospheric scattering models. DehazeNet employs a Convolutional Neural Network (CNN) architecture with specialized layers to embody established assumptions and priors in image dehazing. Specifically, Maxout units are used for feature extraction, and a novel nonlinear activation function called Bilateral Rectified Linear Unit (BReLU) is proposed to improve the quality of the recovered image. The paper establishes connections between DehazeNet components and existing methods, demonstrating superior performance over state-of-the-art techniques while maintaining efficiency and ease of use. Experiments on benchmark images show that DehazeNet outperforms other methods in terms of both quantitative metrics (e.g., MSE, SSIM, PSNR) and qualitative evaluations, particularly in handling challenging scenarios such as sky regions and noise robustness.