A SEGMENTATION METHOD FOR KUZUSHIJI BASED ON K-MEANS CLUSTERING

Ancient Japanese books record a great amount of information, which are valuable research materials in study of history and culture. Over the past few years, there was a large-scale research on digitization of ancient Japanese books and we are convenient to use them due to open access on the Internet. However, it is a challenging problem to recognize those ancient Japanese books due to the complex background and unsteady shape of Japanese characters, which is called Kuzushiji. In this paper, we proposed a method to segment Japanese characters by using image processing and clustering. Our method is based on the analysis of character characteristics, which could identify the segmentation points more accurately. The validity of the proposed method was confirmed by the evaluation experiment.