Abstract

Recently, two new correlation filters have been proposed for distortion-invariant pattern recognition. The first is the minimum variance synthetic discriminant function (MVSDF) in which the output variance is minimized while satisfying the specified correlation plane constraints. In the second technique, known as the minimum average correlation energy (MACE) filter, the average energy in the correlation plane is minimized while satisfying the specified correlation plane constraints. Thus MACE provides very sharp correlation peaks, whereas the MVSDF is the least sensitive to input noise. Since only intensity is usually detected in the output plane, we can introduce arbitrary phases in the correlation plane constraints.

© 1987 Optical Society of America