Object Polarization Information Extraction And Application Under The Aerosol Scattering

Polarization is one of the most useful and important properties of light. It contains not only the traditional light intensity information, but also contains the information that reflects the surface features, the refractive index and so on. Polarization imaging is the common methods measuring the polarized light. However, in the haze weather, due to the scattering of the aerosol in random media we could not obtain object polarization information directly. This is because the measuring light is the synthesis of arilight and the object reflected light, and they are partially polarized light. Therefore, the main content of this thesis is extracting the object polarization information under the atmosphere aerosol and applying it in hazy image restoration. This thesis research includes the follows contents:(1) Object polarization extraction under the atmosphere aerosol. In this thesis, we analyzes the current polarization imaging model’s problem and propose a new model which considering the polarization of the airlight and the object radiance jointly. Based on the assumption that the depth and the radiance of the objects do not exhibit local relation, we proposed a de-correlation method to estimate the polarization degree of object from the polarized images.(2) Reconstruction of the haze-free image based on our polarization hazy imaging model. The current polarization-based dehazing algorithms are based on the assumption that the polarization is only associated with the airlight. This thesis proposed a new polarization imaging model, which considers the polarization of airlight and the object radiance jointly. Our model improved the current polarization filtering method’s applicability and effectiveness.(3) Post-processing of the dehazing results. Because of the current dehazing results is noisy, we propose a blind image denoising algorithm. First, we proposed a fast block-matching and 3D filtering (FBM3D) algorithm. Then a method based on iterator is proposed to estimate the optimum image noise level. The experimental results show the validity of our post-processing algorithm.