Dual Channel night image dehazing algorithm via adaptive SLIC super-pixel Segmentation

Addressing the challenges of reduced brightness, blocking artifacts, and color distortion in night dehazing images due to artificial light source, we propose a dual channel night image dehazing algorithm via adaptive SLIC super-pixel segmentation. Firstly, we introduce an adaptive SLIC super-pixel segmentation algorithm, specifically designed based on image size, for computing the light and dark channel maps, and the improved quadtree algorithm is combined to obtain atmospheric light map that better conforms to the imaging law of night hazy images. Secondly, utilizing the dual channel prior theory, the transmittance of the light channel and the transmittance of the dark channel are estimated separately. Then, the transmittance is weighted using channel differences as weighting coefficients, and guided filtering is used to correct the transmittance. Finally, a threshold limit is imposed on the defogging model to improve the color deviation problem. By comparing with several classical, experiments on synthetic datasets and real world show that our algorithm achieves normal color and moderate brightness in dehazing images, and has an advantage in qualitative and quantitative experiments.