Shape From Shading

A classic problem in computer vision is to compute the shape of an object from its grey-scale image. We're treating the interesting and relatively unexplored case of a terrain under diffuse lighting, as might occur on an overcast day. The input consists of a greyscale image I(x,y) and the output consists of a depth map z(x,y).

This is joint work with Michael Langer of NEC Research.

We make the following assumptions:

• the visible surface is Lambertian with known albedo, rho,
• the depth map is a continuous function of x and y (i.e. it's a terrain),
• the light source is a uniform, diffuse hemisphere over the scene, and
• there are no shadows from objects outside the scene.

More information is in the paper.

Back to James Stewart's home page