Document Structure Analysis Using Back-Propagation Network

摘要

The digitalization of paper-based documents is an important issue in information era. The digitalization means the process of scanning paper documents, analyzing the layout of document image, and converting it into texts by Optical Character Recognition (OCR) technique. Due to that texts recognized by OCR do not contain structural information, digitized document cannot derive hierarchical structure. In this paper, we propose a document structure analysis technique by means of layout information and neural networks. The objective is to classify blocks in a document, such as title blocks and paragraph blocks, into a structural hierarchy and transfer it into an XML (extensible Markup Language) document automatically. Conference paper is chosen in our experiments. In the experiments, we can achieve 94.44% correctness rate using the back-propagation network (BPN) model. Experimental results validate the feasibility of the proposed scheme.