A Document Analysis System Based on Text Line Matching of Multiple OCR Outputs
Conference paper
Abstract
It is well known that integration of multiple OCR outputs can give higher performance than a single OCR. This idea was applied to the printed Japanese recognition and better performance was obtained. In the conventional experiments, however, the zoning, i.e. the extraction of the text region, was done manually and this has been a serious problem from the practical point of view. To solve the problem, an approach to match automatically the classified regions outputted by multiple OCRs was proposed. By the proposed method, a high recognition rate of 98.8% was obtained from OCR systems whose performance is no better than 97.6%.
Keywords
Recognition Rate Document Image Text Region High Recognition Rate Japanese Document
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.
Download
to read the full conference paper text
References
- 1.Baldonado, M., Chang, C.-C.K., Gravano, L., Paepcke, A.: The Stanford Digital Library Metadata Architecture. Int. J. Digit. Libr. 1, 108–121 (1997)CrossRefGoogle Scholar
- 2.Bruce, K.B., Cardelli, L., Pierce, B.C.: Comparing Object Encodings. In: Ito, T., Abadi, M. (eds.) TACS 1997. LNCS, vol. 1281, pp. 415–438. Springer, Heidelberg (1997)CrossRefGoogle Scholar
- 3.van Leeuwen, J. (ed.): Computer Science Today. LNCS, vol. 1000. Springer, Heidelberg (1995)zbMATHGoogle Scholar
- 4.Michalewicz, Z.: Genetic Algorithms + Data Structures = Evolution Programs, 3rd edn. Springer-Verlag, Berlin Heidelberg, New York (1996)zbMATHGoogle Scholar
Copyright information
© Springer-Verlag Berlin Heidelberg 2004