Abstract
Atmospheric pollution by PM2.5 is exactly getting serious at China. The de-hazing or de-fogging task for polluted images has been a long-pending question at NASA remote sensing project. Currently the atmospheric scattering model is a mainstream for recovering the haze-less scene. Above all, a single image de-hazing model based on dark channel prior hypothesis by K. He et al. is most notable in practice. The key to solving this ill-posed problem lies in how to estimate the spatially smooth scene transmittance. This paper proposes an estimation method for spatially smooth scene transmittance based on atmospheric scattering model. The model introduces an anisotropic spatial filtering to reduce the banding artifact as a fatal drawback in the He’s model.The simulation shows how the proposed model enables us to look the clear scene through heavy PM 2.5 pollution or dense fog. Since any small particles are floating in the air even if under clear skies, we never see the hazeless true scenes. So far the de-hazing technologies have mainly targeted at heavy polluted scenes, but may be useful even for usual scenes. The model demonstrates how the brilliant scenic colors are restored from a slight hazy landscape by just adjusting the veiling factor for the scattered air light. Considering that we perceive aerial perspective through the air layer, the estimated scene transmittance must reflect the depth map. Lastly the paper introduces a novel application to the foreground/background separation for a perspective scene and visual effects on the spatial filtering corresponding to the each separated areas.
