Computer Vision research community have been making significant efforts 
towards automatic reconstruction of 3D shape of human face by series of images. 
A number of methods have been proposed for the purpose. 
But none of them is capable of completely automatic reconstruction. 
We propese a method for automatic facial shape recovery by single image or by series of images. 
Our approach is based on the assumption that a set 
of all possible shapes of human face is convex to some extent. 
A shape of particular face is thus approximated by convex combination of some support shapes. 
We call a set of the support shapes a morphable model. 
In order to determine convex combination of the shapes 
one has to establish correspondence between points of the shapes. 
We solve this problem by reducing it to soft constraint satisfaction problem.