We present the theoretical backgrounds of processing the multidimensional data in the case, when the initial matrix has omissions (and values of some particular characteristics are not known because of some reasons). In order to take away uncertainty in initial data, we introduce a special measure of closeness of two vectors with positive components. This measure satisfies practically all the axioms of metrics and, at that, does not depend on the units of measurement. This method is used to represent the generalized index of matrix (vector-column), the values of which should be determined by solution of the optimization problem.