抄録
In this paper, a method for automatic identification of newspaper fonts by template matching is proposed. In template matching between the font images, the genetic algorithm (GA) is used to search the optimum parameters of the translation, the rotation, and the magnification of the template image efficiently. This search processing is executed for many template images, and the template image having the maximum fitness value (the maximum matching rate) is selected as the identification result.
As the result of the identification experiment using two representative Japanese characters of 68 newspapers, the fitness value between the same fonts was higher than that between different fonts. This difference of the fitness values made it possible to specify almost all of these newspaper fonts. When the original font was used, the name of the newspaper could almost be specified. In the case of using the same font in the affiliated publishers, the group name of the publishers could be specified.
In executing the identification program, no complicated setting operation is required. Only by reading in an input image with the scanner and starting the program after trimming the image, the font image is identified automatically. It takes about 2 hours and 50 minutes to execute the font matching for an input image.
