Item

Abstract

An image-based acquisition pipeline for the reflective properties of a human face is presented, proposing an approach which extracts geometry and a spatially varying BRDF of a human face from few measurements in calibrated point light conditions. Applying the morphable face model introduced by Blanz-Vetter for the estimation of geometry in each of the input images, we gain an image-based geometry estimation which does not require range scans of the individual face, and obtain a parameterization of the face surface, which is consistent for all human faces. By means of non-linear optimization, we recover spatially varying parameters for an analytic BRDF model; the resulting BRDF function, together with the estimated geometry, creates a relightable representation of the face. The set of images used for the reflectance measurement is identical to the set of images used for the geometry estimation. Several BRDF models (Cook-Torrance, Lafortune, Ward, and a custom phe\-nom\-e\-no\-logical model) are compared in an experiment.