An important proportion of documents are document images, i.e. scanned documents. For their retrieval, it is important to recognize their contents. Current technologies for optical character recognition (OCR) and document analysis do not handle such documents adequately because of the recognition errors. In this paper, we describe an approach that integrates the detection of errors in scanned texts without relying on a lexicon, and this detection is integrated in the research process. The proposed algorithm consists of two basic steps. In the first step, we apply editing operations on OCR words that generate a collection of error-grams and correction rules. The second step uses query terms, error-grams, and correction rules to create searchable keywords, identify appropriate matching terms, and determine the degree of relevance of retrieved document images. Algorithms has been tested on 979 document images provided by Media-team databases from Washington University, and the experimental results obtained show the effectiveness of our method and indicate improvement in comparison with the standard methods such as exact or partial matching, N-gram overlaps, and Q-gram distance.
Information Retrieval Based on OCR Errors in Scanned Documents
2003-06-01
102303 byte
Conference paper
Electronic Resource
English
Compression and decompression of scanned technical documents
Automotive engineering | 1993
|Text, photo, and line extraction in scanned documents
British Library Online Contents | 2012
|93ME001 Compression and Decompression of Scanned Technical Documents
British Library Conference Proceedings | 1993
|Method for estimating lineature by scanned samples of documents
British Library Online Contents | 2009
|Pre- and postprocessing for multilayer compression of scanned documents
British Library Online Contents | 2011
|