Computer recognition of 2-D patterns using generalized matched filters

Prior work on generalized matched filters (GMF(s)) was limited to eighty points and a 1-D signal. We show here how to calculate arbitrarily large GMF(s). We then compare GMF(s) with other common pattern recognition filters for one particular image using computer simulation. The GMF appears to offer improvements in terms of smaller within-class variability and greater between-class separation relative to matched filters. In addition, the GMF permits far more accurate location of the object in the input scene than the matched filter.