Weighted sparse representation for human ear recognition based on local descriptor