Estimating 3-D body parameters from reflection component separating imagery: a performance analysis

Iterative least-squares estimation requires accurate reflectance models to retrieve geometrical parameters of 3-D objects from an image projection. We investigate the use of separating the diffuse (body) reflection from the specular (surface) reflection, where the latter is responsible for image highlights. The performance of several models has been analysed by comparing local higher-order derivatives of the least-squares error function. Experiments show that the (smooth) diffuse component yields the best convergence properties, while the (sharp) specular component cast be utilized to improve noise insensitivity.