Text this: On combining multiple features for cartoon character retrieval and clip synthesis