Dehazing of four-channel remote sensing images fused with near-infrared

The paper presented an enhanced dark channel prior (DCP) technique for removing haze from remote sensing images. Unlike traditional methods that use only the red, green, and blue bands, the proposed approach integrates the near-infrared (NIR) band, exploiting its superior haze-penetrating capability and the complementary information it offers with visible light bands to improve dehazing outcomes. Traditional single-image dehazing techniques, such as those based on atmospheric scattering models, often depend on assumptions and prior knowledge about the scene, which may not be valid in complex environments, resulting in subpar dehazing results and the introduction of artifacts. Deep learning methods, in contrast, are heavily reliant on vast amounts of training data and face challenges in ensuring generalization across various remote sensing images, and may also be impacted by information loss during image compression. This study introduces an improved dark channel calculation method that incorporates information from all four bands (R, G, B, and NIR), circumventing the direct use of NIR band fusion for color distortion correction and utilizing NIR information to enhance dehazing performance. Experimental results with real remote sensing datasets show that the proposed method effectively reduces haze and enhances image clarity, and has the potential to preserve the spectral integrity across all four bands.