Text extraction via an edge-bounded averaging and a parametric character model.