Dealing with outliers in high dimensional data: a COMALS procedure