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

Abstract

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.

How well does multiple OCR error correction generalize?

“…Previous work 2,13,26 using the Eisenhower Communiqués test set and the Enron training set was focused on individual document performance in which corpus WERs were calculated as the average of the individual document WERs. A result of this method is that corpus statistics would give more weight to the the tokens of a short document than to the tokens of a long document.…”

“…Our previous work 13,26 showed the improvement in the WER using a trained machine learner with the alignment of the output from multiple OCR engines. Reinterpreting these previous results using micro-averaging techniques, the underlined entries in Table 4 show the decreasing WER as additional OCR outputs are added to the alignment.…”

“…23 Esakov, Lopresti, and Sandberg 24 evaluated recognition errors of OCR systems, and Kolak, Byrne, and Resnik 25 specifically applied their algorithms to OCR systems for post-OCR error correction in natural language processing tasks. OCR error correction with in-domain training 26 as well as out-of-domain training using a synthetic training dataset 13 have been shown to be effective. Recent work by Yamazoe et al 27 effectively uses multiple weighted finite-state transducers (WFST) with both the OCR and a lexicon of the target language(s) to resolve the ambiguity inherent in line-and character-segmentation, and character recognition, in which the number of combinations can be very large.…”

On Automating Editions

“…19 Despite this warning, since the early 2010s the text recognition community has said that "the OCR of modern clean documents is e ectively a solved problem" although "older degraded documents present di culties." 20 HTR is mentioned only in passing in Digital Scholarly Editing: Theories and Practices as research that "has not yet produced reliably working products, but much more is to be expected in the coming years." 21 However, there was an implication that digital editions that depend on any such automated processes are not "proper," 22 which may speak to issues of scholarly ownership of the entire editing process, given that the mechanisms of many OCR and HTR technologies remain black-box.…”

OCR Training and Simulation for Person Follower Robot in an Indoor Environment

“…A degradation model was proposed and analyzed by [13] to recognize the similarity between groups of ruined characters. The authors of [14] proposed a method of reducing word error rate using different OCR engines and the usage of in domain training was explored. An iterative training framework was proposed by [15] which uses OCR without segmentation to reduce the error in character recognition.…”