Visual pattern recognition using morphological methods
Ioannis N. M. Papadakis, James G. Reisman, Stelios C. A. Thomopoulos · 1992
An effective method for visual pattern recognition using morphological techniques is presented. It is shown that it can be successfully used for the recognition of deformed letters. The method extracts morphological information by the successive dilation of an idealized letter set. At each stage a properly defined similarity index is computed. The maximum values of the similarity index for each stored letter are compared and are used for the classification decision. Classification results are presented for a set of deformed capital English letters with a realistic level of deformation. Skeletonization of the deformed pattern is shown to improve the performance of the classification method. The described method can be easily implemented using a parallel architecture.>