OCR quality: Key to enhanced Data Mining

In this paper we present a simple yet effective approach to automatic OCR error detection and correction on a corpus of French clinical reports of variable OCR quality within the domain of foetopathology. While traditional OCR error detection and correction systems rely heavily on external information such as domain-specific lexicons, OCR process information or manually corrected training material, these are not always available given the constraints placed on using medical corpora. We therefore propose a novel method that only needs a representative corpus of acceptable OCR quality in order to train models. Our method uses recurrent neural networks (RNNs) to model sequential information on character level for a given medical text corpus. By inserting noise during the training process we can simultaneously learn the underlying (character-level) language model and as well as learning to detect and eliminate random noise from the textual input. The resulting models are robust to the variability of OCR quality but do not require additional, external information such as lexicons. We compare two different ways of injecting noise into the training process and evaluate our models on a manually corrected data set. We find that the best performing system achieves a 73% accuracy.

Digital collections are increasingly used for a variety of purposes. In Europe only, we can conservatively estimate that tens of thousands of users consult digital libraries daily. The usages are often motivated by qualitative and quantitative research. However, caution must be advised as most digitized documents are indexed through their OCRed version, which is far from perfect, especially for ancient documents. In this paper, we aim to estimate the impact of OCR errors on the use of a major online platform: The Gallica digital library from the National Library of France. It accounts for more than 100M OCRed documents and receives 80M search queries every year. In this context, we introduce two main contributions. First, an original corpus of OCRed documents composed of 12M characters along with the corresponding gold standard is presented and provided, with an equal share of English- and French-written documents. Next, statistics on OCR errors have been computed thanks to a novel alignment method introduced in this paper. Making use of all the user queries submitted to the Gallica portal over 4 months, we take advantage of our error model to propose an indicator for predicting the relative risk that queried terms mismatch targeted resources due to OCR errors, underlining the critical extent to which OCR quality impacts on digital library access.

In digital libraries, the accessibility of digitized documents is directly related to the way they are indexed. Named entities are one of the main entry points used to search and retrieve digital documents. However, most digitized documents are indexed through their OCRed version and OCR errors may hinder their accessibility. This paper aims to quantitatively estimate the impact of OCR quality on the performance of named entity recognition (NER). We tested state-of-the-art NER techniques over several evaluation benchmarks, and experimented with various levels and types of synthesised OCR noise so as to estimate the impact of OCR noise on NER performance. We share all corresponding datasets. To the best of our knowledge, no other research work has systematically studied the impact of OCR on named entity recognition over datasets in multiple languages. The final outcome of this study is an evaluation over historical newspaper data of the national library of Finland, resulting in an increase of around 11% points in terms of F1-measure over the best-known results to this day.

We aim to investigate the performance of current OCR systems on low resource languages and low resource scripts. We introduce and make publicly available a novel benchmark, OCR4MT, consisting of real and synthetic data, enriched with noise, for 60 low-resource languages in low resource scripts. We evaluate state-of-the-art OCR systems on our benchmark and analyse most common errors. We show that OCR monolingual data is a valuable resource that can increase performance of Machine Translation models, when used in backtranslation. We then perform an ablation study to investigate how OCR errors impact Machine Translation performance and determine what is the minimum level of OCR quality needed for the monolingual data to be useful for Machine Translation.

In this paper, we evaluate the influence of detection quality on OCR accuracy in automatic license plate recognition (ALPR) systems. Using a modular pipeline, we compare the performance of three OCR engines - Tesseract, EasyOCR, and PaddleOCR -when fed with outputs from two variants of YOLOv8n object detectors: one autotuned using Optuna and another manually fine-tuned. The fine-tuned YOLOv8n model, optimized for Indian license plates, achieved the best results, allowing PaddleOCR to reach an OCR accuracy of 99 %, marking a 4.16 % improvement over the baseline autotuned model. We use a range of evaluation metrics, including Character Error Rate (CER), Word Error Rate (WER), and Normalized Levenshtein Distance (NLD), to comprehensively quantify OCR performance under different detection scenarios. Our findings highlight the critical role of accurate detection in downstream OCR quality and offer valuable insights for building robust, end-to-end ALPR systems, particularly in real-world deployment scenarios.

Crowdsourcing approaches for post-correction of OCR output (Optical Character Recognition) have been successfully applied to several historic text collections. We report on our crowd-correction platform Kokos, which we built to improve the OCR quality of the digitized yearbooks of the Swiss Alpine Club (SAC) from the 19th century. This multilingual heritage corpus consists of Alpine texts mainly written in German and French, all typeset in Antiqua font. Finding and engaging volunteers for correcting large amounts of pages into high quality text requires a carefully designed user interface, an easy-to-use workflow, and continuous efforts for keeping the participants motivated. More than 180,000 characters on about 21,000 pages were corrected by volunteers in about 7 months, achieving an OCR gold standard with a systematically evaluated accuracy of 99.7% on the word level. The crowdsourced OCR gold standard and the corresponding original OCR recognition results from Abbyy FineReader 7 for each page are available as a resource. Additionally, the scanned images (300 dpi) of all pages are included in order to facilitate tests with other OCR software.