An input pattern weighting method is proposed for increasing the efficiency of the recognition of high-correlated patterns. Numerical simulation results are given. An optical implementation of the high-order associative memory model using an inner product scheme with the weighting method is suggested.