Curved surface orientation from shading and surface style indication

This paper develops a new algorithm for extracting 3D surface normal from 2D intensity image, specified for integration of stereo vision and shape from shading. Every curved surface patch is classified into one of four types: dome, cup. horizontal saddle and vertical saddle, according to its signs of surface horizontal and vertical curvatures. And the surface patch style could be derived from the sparse depth set of stereo vision. Four basic operators are designed individually to extract the surface orientation corresponding to each kind of surface patch style. The algorithm is based on Lambertian brightness equation and the orthogonality between surface normal and its tangent vectors. A local optimal method is used to ensure the brightness consistency. It is shown that the new algorithm performs well at the patches where their surface style matches the type of the applied operator.