We present a nonlinear correlation to improve the selectivity for optical p
attern recognition. The approach is based on morphological correlation whic
h involves a threshold decomposition concept. Hereby, we propose a subband
decomposition in the Fourier domain to perform the threshold decomposition
operation. We consider two frequency bands that give rise to two separate c
hannels. We apply the morphological correlation to each channel using a loc
alized threshold decomposition. Then, we define a two-channel morphological
correlation. The final detection decision is made as a combination of both
correlation outputs. The two-channel morphological correlation yields impr
oved discrimination capability and noise robustness compared with common li
near and morphological correlation. Computer simulations are presented as w
ell as optical experiments performed using a joint transform correlator.