Multichannel pattern recognition based on the projection of the components of a multichannel image onto a vector followed by single-channel matched filtering is presented. The projection preprocessing reduces the multichannel information to a single channel, while most of the necessary information is maintained. We describe the design of an optimal filter in terms of robustness to additive noise. A phase-only filter obtained on the basis of the optimal filter can be used for optical pattern recognition. One advantage of the phase-only filter described here is its improved robustness to noise. Computer simulation results are provided to illustrate multichannel pattern recognition using the proposed method.