Haze-level prior approach to enhance object visibility under atmospheric degradation

Abstract

Outdoor captured scenes are degraded by atmospheric particles and water droplets. Due to scattering and absorption effects in the atmosphere, degraded images lose contrast and color fidelity. Performances of the computer vision algorithms are bound to suffer from low-contrast scene radiance. In many single-image dehazing models, the larger the deviation in estimation of the key parameters such as transmission map and atmospheric light, the higher the halo artifacts and loss of fine details in the dehazed image. The available models assume that the scattering light is independent of wavelength, as the size of the atmospheric particles is larger compared to the wavelength of light. The model presented in this paper emphasizes the appropriate estimation of intensified transmission map from the hazy images by exploiting the scattering coefficient in order to address the issues of haze concentrations. Experiments conducted on thick and thin hazy images provide an optimal estimation of the model parameters, which can be applied directly in real-time situations. The available models are observed to be inconsistent sometimes in the enhancement of contrast, saturation and color information either together or independently. The proposed model addresses these issues by extracting the haze-relevant features from the hazy images, such as hue disparity, contrast, and darkness, which yield more vivid saturation results. Moreover, the proposed model addresses different haze densities in the scene without the use of refinement filters.