A dehazing method for sea fog images based on pixel-level skylight polarization degree estimation

Sea fog blurs images collected by USV. Conventional polarization-based dehazing methods assume a constant polarization degree of skylight. However, sea fog images exhibit significant variations in scene depth, which does not satisfy this assumption, leading to poor results. Therefore, a dehazing method for sea fog images based on pixel-level skylight polarization degree estimation is proposed. Firstly, Gaussian filtering is employed to estimate the polarization component of skylight for each pixel at different polarization angles. Subsequently, based on the Stokes vectors, the skylight polarization degree of each pixel is calculated. Furthermore, a filtering strategy combining median filtering and gradient domain guided filtering is designed to suppress noise while preserving edge information. Finally, the dehazed images are obtained based on the polarization hazy imaging model. Experimental results show that, compared to traditional methods, the average gradient of dehazed images has been improved by at least 20.22%, validating the effectiveness of the proposed approach.