Machine Segmentation of Unformatted Characters.
Robin A Simmons · 1981
Abstract : This thesis presents a method of segmenting unformatted alpha-numeric characters. Reconstructing the magnitude of the Fourier transform of a template character with the phase of a string of unformatted characters containing the template character causes all characters that do not have the magnitude of the template to be attenuated in the visual domain. The template character will not be attenuated, since it has both proper magnitude and phase, and a peak detector can find the most probable location of the character. This process also gives a first choice for what the character is (the template). Results are presented for most of the English alphabet. (Author)