Image Analysis for Historical Japanese Book Archives

Abstract

This paper describes methods of image analysis for historical Japanese book archives with a dominant focus on character segmentation. The segmentation methodology includes stain and smear removal, binarization, character line extraction, and character extraction by region labeling with integration and separation techniques. The experimental results show that the proposed method can segment all text lines correctly and can extract more than 79% of the characters from 16 pages of Chinsetsu Yumiharizuki, containing 176 text lines and a total of 5181 quite complicated characters.