Bayesian Reflectance Component Separation

We work on a Bayesian approach to the estimation of the specular component of a color image, based on the Dichromatic Reflection Model (DRM). The separation of diffuse and specular components is important for color image segmentation, to allow the segmentation algorithms to work on the best estimation of the reflectance of the scene. In this work we postulate a prior and likelihood energies that model the reflectance estimation process. Minimization of the posterior energy gives the desired reflectance estimation. The approach includes the illumination color normalization and the computation of a specular free image to test the pure diffuse reflection hypothesis.