Experimental optical results with minimum Euclidean distance optimal filters, coupled modulators, and quadratic metrics

Optical correlators used in pattern recognition should have their operation optimized according to physically measurable behavior related to the correlator's task. The optimization is best done with explicit consideration of the limited set of complex values available to the filter. We earlier developed a theoretical filter computing strategy, MEDOF (minimum Euclidean distance optimal filter), that optimizes a wide variety of correlation metrics on physically realistic modulators. We use a single reference image and add noise according to a stated model. We show laboratory results for filters optimizing a variety of metrics on a highly coupled SLM. The results show improvements in the metrics by using MEDOF when compared with some other filter computing strategies such as matching phase. Optimizing SNR appears preferable to the other metrics when detecting an object in clutter of known power spectral density. (C) 1999 Society of Photo-Optical instrumentation Engineers. [S0091-3286(99)01802-4].