Genetic algorithm for thinning gray-scale images of characters

Abstract

In this paper, we propose a new method for thinning digital gray-scale pictures by fitting piece-wise linear line to skelton. The method is not based on the point-wise gray-value information as Salari and Siy method, but based on an objective function robust for noise contaminated pictures. To optimize the function, we search the solution of the optimization problem with genetic algorithm. The procedure of the method is as follows. We allocate small regions so that they cover the neighborhood of skelton of object. In each small region, we extract a linear line which approximates the skelton in the small region, and represent the line by a couple of label numbers of points with nonzero gray values through which the line passes. A string is constructed by collecting all the label pairs. The fitness of a string is given by an objective function which is constructed so as to evaluate the proximity to skelton and to be robust for noise. After a genetic algorithm generates successive populations of strings, a skelton is given by the string with the greatest fitness.