| Abstract: | Conventional holographic pattern recognition systems suffer from the disadvantage that, in dealing with highly ordered patterns such as Arabic numerals or the letters of the Roman alphabet it is sometimes impossible to distinguish between patterns with a high degree of similarity (eg the capital letters ‘E’ and ‘F’ or ‘B’ and ‘P’).The present paper analyzes a simple case and shows how the insertion of a suitable spatial filter in the Fourier transform plane can reduce the unwanted cross-correlation signal between dissimilar patterns to zero, without removing the desired auto-correlation signal between similar patterns. Although no attempt is made to generalize the analysis, it seems probable that in cases involving the recognition of a small number of patterns with a strong ‘family resemblance’ a compromise filtering arrangement could be developed which would lead to a substantially improved performance. |
