Height estimation from vector fields of surface normals

We consider the problem of recovering the height information from a field of surface normals defined as three-dimensional (3-D) vectors. The algorithm starts from a certain location on the lattice and updates iteratively the heights in a neighbourhood. Then it proceeds to the neighbours of the already calculated heights. The updating rule is based on a gradient descent minimising the vector difference between the given surface normal, and the surface normal, extracted from the estimated height map at each site. The proposed algorithm is applied for recovering height information from synthetic aperture radar (SAR) images of the terrain.