http://www.mrl.nyu.edu/~hertzman/hertzmann-intro3d.pdf 
 
The method produced in this paper is almost straight forward, and can be viewed as marching triangle algorithm. We divide the image into triangular meshes and detect the contours which satisfy the condition the the dot product of the normal vector and view vector equals 0(for orthogonal projection). The paper also talks about the visiblity of the surface. 
 
Advantages: Easy to program