| The estimation of complex refraction index and surface geometries (microfacet orientation and surface roughness) from optical scatter off a target's surfaces are tasks of pivotal importance for remote sensing and computer graphics applications such as shape extraction, surface reconstruction and target recognition. However, these estimations often rely on a microfacet polarimetric bidirectional reflectance distribution function (pBRDF) and/or a microfacet degree of polarization (DOP) model that are limited to specular targets involving only single surface scattering. If diffuse scattering is present, the measured polarization dependence will be "contaminated" and depart from the ideal surface reflection response, and therefore, the estimated results can be inaccurate.;In this dissertation, to account for the diffuse scattering component of non-specular materials, we follow theories developed by Kubelka-Munk (1948) and Le Hors et al. (2001) and propose modified pBRDF and DOP models by adding a diffuse component to a microfacet model description. Our approach is general enough that other classical scattering theories, for example by Oren-Nayar-Wolff or Beckmann-Kirchhoff, can be incorporated in the microfacet pBRDF framework. Evaluation of these modified models indicates that diffuse scattering can significantly affect the values of the pBRDF, DOP and Stokes parameters, and these biased values can further lead to inaccurate estimation of refractive index and surface roughness.;In addition, we find that coupling these modified models with a nonlinear equation solver leads to new approaches for jointly estimating refractive index and surface roughness from real measurements. When applied to laboratory-measured values, these modified models produce significantly improved estimated results relative to reference values; and synthesized results using the estimated parameters show improved correspondence with measurements. By accurately estimating the complex refractive index, surface roughness parameter and diffuse scattering coefficient for each pixel of a polarimetric image, parameter-based images are formed, which provide a useful means to identify and extract the targets. |
