Visual Realism in Digital Heritage

To be convincing, digital heritage needs to represent physical objects and render them in a way that looks realistic when viewed on a display screen. The problem is difficult enough for flat 2D objects viewed under arbitrary sources of illumination, but is much harder for 3D objects with curved surfaces of diverse materials, especially with varying degrees of sheen or gloss. A new method is presented for estimating the reflectance distribution function from a set of images taken in an illumination dome. Surface normals and albedo are calculated by regression over a subset of the intensity at each pixel. The ratio of actual intensity to diffuse intensity enables the direction of peak specularity to be determined. Then the distribution of specular intensity is modelled as a modified Lorentzian function of radial angle from the specular peak direction, with circular cross section for isotropic surfaces and elliptical cross section for anisotropic surfaces. Fitting the results over the image enables a characteristic RDF (a two-dimensional subset of the BRDF for fixed viewing direction) to be estimated. Rendering the surface by a model that adds the body colour (diffuse) and specular (glistening) components at each pixel for any direction of incident illumination is shown to provide a good simulation of the appearance of the original object.