Improving optical character recognition through efficient multiple system alignment

Lund, William B.; Ringger, Eric K.

doi:10.1145/1555400.1555437

Cited by 19 publications

(11 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The texts from the three OCR engines are character aligned using the A* algorithm with the Reverse Dijkstra admissible heuristic described by Lund and Ringger [8]. From this character level alignment we construct a lattice of word hypotheses such that wherever there is agreement across all engines on the location of white space we construct a column of hypotheses.…”

Section: A Baseline Ocr Resultsmentioning

confidence: 99%

“…Our research leverages the variation among OCR engines (see Figure 1) and additional features of the OCR hypotheses to improve the output beyond what any single OCR engine is capable of. In this case, where in-domain training data is available, we improve upon our previous work [8] and show how a decision list trained on in-domain data using feature combinations reduces the word error rate beyond what is achieved using consensus voting or dictionary matching alone. Further, we explore using a spell checker to suggest additional words for hypotheses that do not appear in the dictionary or gazetteers..…”

Section: Introductionmentioning

confidence: 84%

“…The National Library of Medicine [16] selected Prime OCR, a commercial system that votes on responses from multiple OCR engines, to improve word recognition. The authors' previous work [8] also explored voting in conjunction with feature engineering. The approach reported in this work uses distinct OCR engines to find variations in the text recognized by the systems, and similar to Lopresti, we find that voting plays an important role in making good corrections.…”

Section: B Related Workmentioning

confidence: 99%

See 2 more Smart Citations

Error Correction with In-domain Training across Multiple OCR System Outputs

Lund

Ringger

2011

2011 International Conference on Document Analysis and Recognition

Self Cite

View full text Add to dashboard Cite

Optical character recognition (OCR) systems differ in the types of errors they make, particularly in recognizing characters from degraded or poor quality documents. The problem is how to correct these OCR errors, which is the first step toward more effective use of the documents in digital libraries. This paper demonstrates the degree to which the word error rate (WER) can be reduced using a decision list on a combination of textual features across the aligned output of multiple OCR engines where in-domain training data is available. This research was performed on a data set for which the mean WER across the three OCR engines employed is 33.5%, and the lattice word error rate is 13.0%. Our correction method leads to a 52.2% relative decrease in the mean WER and a 19.5% relative improvement over the best single OCR engine, as well as an improvement over our previous work. Further, our method yields instances where the document WER approaches and for five documents matches the lattice word error rate, which is a theoretical lower bound given the evidence found in the OCR.

show abstract

Section: A Baseline Ocr Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 84%

Section: B Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Error Correction with In-domain Training across Multiple OCR System Outputs

Lund

Ringger

2011

2011 International Conference on Document Analysis and Recognition

Self Cite

View full text Add to dashboard Cite

show abstract

“…Ringlstetter et al [27] suggested a method to discriminate character confusions in multilingual texts. Cecotti et al [6] and Lund and Ringger [16] aligned multiple OCR outputs and illustrated strategies for selection. Namboodiri et al [20] and Zhuang and Zhu [32] integrated multi-knowledge with the OCR output in post-processing, such as fixed poetical structures for Indian poetry or semantic lexicons for Chinese texts.…”

Section: Related Workmentioning

confidence: 99%

Improving OCR Accuracy for Classical Critical Editions

Boschetti

Romanello

Babeu

et al. 2009

Research and Advanced Technology for Digital Libraries

View full text Add to dashboard Cite

Abstract. This paper describes a work-flow designed to populate a digital library of ancient Greek critical editions with highly accurate OCR scanned text. While the most recently available OCR engines are now able after suitable training to deal with the polytonic Greek fonts used in 19th and 20th century editions, further improvements can also be achieved with postprocessing. In particular, the progressive multiple alignment method applied to different OCR outputs based on the same images is discussed in this paper.

show abstract

“…Similarly, Optical Character Recognition (OCR) engines, despite great success in recent attempts such as Google Books or Internet Archive, are not without problems, and often produce error-abundant text data. [13] pointed out that although researchers are having increasing levels of success in digitizing hand-written manuscripts, Table 1: Top words (selected by LDA) for five topics of a small sample of Unlv OCR data set (erroneous words are in italic).…”

Section: Introductionmentioning

confidence: 99%

On handling textual errors in latent document modeling

Yang

Lee

2013

Proceedings of the 22nd ACM International Conference on Information &Amp; Knowledge Management

View full text Add to dashboard Cite

As large-scale text data become available on the Web, textual errors in a corpus are often inevitable (e.g., digitizing historic documents). Due to the calculation of frequencies of words, however, such textual errors can significantly impact the accuracy of statistical models such as the popular Latent Dirichlet Allocation (LDA) model. To address such an issue, in this paper, we propose two novel extensions to LDA (i.e., TE-LDA and TDE-LDA): (1) The TE-LDA model incorporates textual errors into term generation process; and (2) The TDE-LDA model extends TE-LDA further by taking into account topic dependency to leverage on semantic connections among consecutive words even if parts are typos. Using both real and synthetic data sets with varying degrees of "errors", our TDE-LDA model outperforms: (1) the traditional LDA model by 16%-39% (real) and 20%-63% (synthetic); and (2) the state-of-the-art N-Grams model by 11%-27% (real) and 16%-54% (synthetic).

show abstract

Improving optical character recognition through efficient multiple system alignment

Cited by 19 publications

References 13 publications

Error Correction with In-domain Training across Multiple OCR System Outputs

Error Correction with In-domain Training across Multiple OCR System Outputs

Improving OCR Accuracy for Classical Critical Editions

On handling textual errors in latent document modeling

Contact Info

Product

Resources

About