Intelligent parsing of scanned volumes for web based archives

The proliferation of digital libraries and the large amount of existing documents raise important issues in efficient handling of documents. Printed texts in documents need to be converted into digital format and semantic information need to be parsed and managed for effective retrieval. In this work, we attempt to solve the problems faced by current web based archives, where large scale repositories of electronic resources have been built from scanned volumes. Specifically, we focus on the scientific domain and target scanned volumes of scientific publications. Our goal is to automate the semantic processing of scanned volumes, an important and challenging step towards efficient retrieval of content within scanned volumes. We tackle the problem by designing a machine learning-based method to extract multi-level metadata about content of scanned volumes. We combine image and text information within scanned volumes for intelligent parsing. We developed a system and test it with real world data from the Internet Archive, and the experimental evaluation has demonstrated good results.