New procedures controlling the false discovery proportion via Romano-Wolf's heuristic