Word Segmentation Algorithm Based on Recognition of Letter Features

摘要

The purpose of this thesis was to develop a computer algorithm to segment letters in a word so that pattern recognition techniques such as two dimensional Fourier Transforms could be used to recognize the individual letters. This study did not intend to cover various pattern recognition techniques but, as the algorithms developed letter features were recognized to gather information or clues on adjacent touching letters to decide on possible segmentation locations. This study was limited to the segmentation of lower case letters in the English alphabet excluding stylized print and italic print. Lower case print was chosen because it represents the worst case task for a segmentation algorithm. Two adjacent lower case letters often look like a third lower case letter. However in upper case letters similar occurrences are rare. Hand printed touching letters were selected to demonstrate the validity of the algorithm. Keywords: reading machines; character recognition. (Author)