next up previous contents
Next: ShapesShocks, and Deformations Up: Computer Vision Previous: Massively Parallel Radiosity in

Recovering Shape from Shading and Contour

P. Breton, S.W. Zucker The classical approach to shape from shading problems is to find a numerical solution of the image irradiance partial differential equation. It is usually assumed that the parameters of this equation (the light source direction and surface albedo) can be estimated in advance. For images which contain shadows and occluding contours, this decoupling of problems is artificial and the classical approach fails. We develop a new approach to solving these equations using the image geometric structures instead of the image photometric structure. Our approach is based on modern differential geometry, and solves for light source and surface shape changes concurrently. Local scene elements (scenels) are estimated from the shading flow field and the tangent field (see Fig. 10), and smoothness, material, and light source compatibility conditions resolve them into consistent scene descriptions. Shadows and related difficulties for the classical approach are discussed. [].40 The shading flow field and the tangent field are geometrical structures that characterize respectively, smooth and abrupt changes in image intensity. These geometrical structures were extracted from the image of a sphere illuminated from two distant light sources: the shading flow field is depicted by thin arrows, and the tangent field, by thick arrows. zucker:SphereMLsff

Thierry Baron
Mon Apr 7 12:54:24 EDT 1997