Reconstructing Scanned Documents for Full-text Indexing to Empower Digital Library Services

Melania Nitu, Mihai Dascalu, Maria-Iuliana Dascalu, Teodor-Mihai Cotet and Silvia Tomescu

The digital era raises new challenges for traditional library services in which information has to be delivered and supported by technology-enhanced systems. The increasing need for rapid access to information requires librarians to re-evaluate the way they develop, manage and deliver resources, as well as services. However, most information extraction systems are not designed to work with PDF files generated after Optical Character Recognition, and several problems are encountered while trying to properly restructure the recognized text, for example: disruption of paragraphs, improper page breaks, or loss of content structure. This paper introduces a pre-processing pipeline designed to support university libraries to adequately index old document collections. The extracted text is indexed into Elasticsearch which facilitates the search for relevant documents, based on keywords. The information extraction system is designed to assist librarians in the digitization process by enabling a systematic review of documents, which leads to more accurate representations of the indexed files.