Underline removal method by utilizing characteristics of Japanese business documents

In this paper we propose an underline removal method specific to Japanese business document. Automated removal of underlines is important, because underline is the main cause of OCR misrecognition. The main feature of our method is to remove various types of underlines such as touched, inclined, and blurred lines by line template matching. Moreover, our method makes it possible to remove all possible underlines by excluding table ruled lines which are necessary for document structure analysis. The experimental result demonstrates that the proposed method is able to improve OCR recognition accuracy.