In cases in which the image-to-image spatial variability of the input pattern changes with the spatial location, a localized-filtering method should be used for pattern recognition. Localized space-invariant filtering is investigated, and its improved recognition abilities are demonstrated with the recognition of fingerprints. The motivation for the investigated implementation is related to the fact that a person never presses his finger on a surface with equal pressure. This variation results in different amounts of spatial shifting being required from the optical processor in different regions of the fingerprint. A two-region mathematical model for representing the human finger is presented and investigated by use of localized space-invariant filtering by means of a computer.