Text this: Learning word and scene semantics for classification and retrieval of complex images /