Pattern recognition by quantization of modulated function to complexes of a quadrupole in matched filtering

An improved matched filtering method for pattern recognition is proposed. In this method, functions modulated by the matched filter are quantized to the complexes of a quadrupole in the frequency domain. It is shown that shaper autocorrelation and much more suppressed cross-correlation signals can be obtained over those achieved by ordinary matched filters. It is also demonstrated that video-rate correlation processing with an 256×256 input image size can be performed on a regular personal computer using the proposed method even when the reference pattern is switched for every input image.