Optical Character Recognition Text Research Articles

A Hybrid Multimodal Data Fusion-Based Method for Identifying Gambling Websites

With the development of network technology, the number of gambling websites has grown dramatically, causing a threat to social stability. There are many machine learning-based methods are proposed to identify gambling websites by analyzing the URL, the text, and the images of the websites. Nevertheless, most of the existing methods ignore one important piece of information, i.e., the text within the website images. Only the visual features of images are extracted for detection, while the semantic features of texts on the images are ignored. However, these texts have key information clearly pointing to gambling websites, which can help us identify such websites more accurately. Therefore, how to fuse image and text multimodal data is a challenge that should be met.Motivated by this, in this paper, we propose a hybrid multimodal data fusion-based method for identifying gambling websites by extracting and fusing visual and semantic features of the website screenshots. First, we fine tune the pretrained ResNet34 model to train an image classifier and to extract visual features of webpage screenshots. Second, we extract textual content from webpage screenshots through the optical character recognition (OCR) technique. We use pretrained Word2Vec word vectors as the initial embedding layer and use Bi-LSTM to train a text classifier and extract semantic features of textual content on the screenshots. Third, we use self-attention to fuse the visual and semantic features and train a multimodal classifier. The prediction results of image, text, and multimodal classifiers are fused by the late fusion method to obtain the final prediction result. To demonstrate the effectiveness of the proposed method, we conduct experiments on the webpage screenshot dataset we collected. The experimental results indicate that OCR text on the webpage screenshots has strong semantic features and the proposed hybrid multimodal data fusion based method can effectively improve the performance in identifying gambling websites, with accuracy, precision, recall, and F1-score all over 99%.

PNRank: Unsupervised ranking of person name entities from noisy OCR text

Text databases have grown tremendously in number, size, and volume over the last few decades. Optical Character Recognition (OCR) software is used to scan the text and make them available in online repositories. The OCR transcription process is often not accurate resulting in large volumes of garbled text in the repositories. Spell correction and other post-processing of OCR text often prove to be very expensive and time-consuming. While it is possible to rely on the OCR model to assess the quality of text in a corpus, many natural language processing and information retrieval tasks prefer the extrinsic evaluation of the effect of noise on the task at hand. This paper examines the effect of noise on the unsupervised ranking of person name entities by first populating a list of person names using an out-of-the-box Named Entity Recognition (NER) software, extracting content-based features for the identified entities, and ranking them using a novel unsupervised Kernel Density Estimation (KDE) based ranking algorithm. This generative model has the ability to learn rankings using the data distribution and therefore requires limited manual intervention. Empirical results are presented on a carefully curated parallel corpus of OCR and clean text and “in the wild” using a large real-world corpus. Experiments on the parallel corpus reveals that even with a reasonable degree of noise in the dataset, it is possible to generate ranked lists using the KDE algorithm with a high degree of precision and recall. Furthermore, since the KDE algorithm has comparable performance to state-of-the-art unsupervised rankers, using it on real-world corpora is feasible. The paper concludes by reflecting on other methods for enhancing the performance of the unsupervised algorithm on OCR text such as cleaning entity names, disambiguating names concatenated to one another and correcting OCR errors that are statistically significant in the corpus.

Simple is not Easy: A Simple Strong Baseline for TextVQA and TextCaps

Texts appearing in daily scenes that can be recognized by OCR (Optical Character Recognition) tools contain significant information, such as street name, product brand and prices. Two tasks -- text-based visual question answering and text-based image captioning, with a text extension from existing vision-language applications, are catching on rapidly. To address these problems, many sophisticated multi-modality encoding frameworks (such as heterogeneous graph structure) are being used. In this paper, we argue that a simple attention mechanism can do the same or even better job without any bells and whistles. Under this mechanism, we simply split OCR token features into separate visual- and linguistic-attention branches, and send them to a popular Transformer decoder to generate answers or captions. Surprisingly, we find this simple baseline model is rather strong -- it consistently outperforms state-of-the-art (SOTA) models on two popular benchmarks, TextVQA and all three tasks of ST-VQA, although these SOTA models use far more complex encoding mechanisms. Transferring it to text-based image captioning, we also surpass the TextCaps Challenge 2020 winner. We wish this work to set the new baseline for these two OCR text related applications and to inspire new thinking of multi-modality encoder design. Code is available at https://github.com/ZephyrZhuQi/ssbaseline

Infrastructure for Long-term Preservation and OCR Analysis of Herbarium Images

Herbaria hold large numbers of specimens: approximately 22 million herbarium specimens exist as botanical reference objects in Germany, 20 million in France and about 500 million worldwide. High resolution digital images of these specimens take up substantial bandwidth and disk space. New methods of extracting information from the specimen labels have been developed using OCR (Optical character recognition) techniques, but the exploitation of this technology for biological specimens is particularly complex due to the presence of biological material in the image with the text, the non-standard vocabularies, alongside the variation and age of the fonts. Much of the information is handwritten and natural handwriting pattern recognition is a less mature technology than OCR. Today, our system (eTDR-European Trusted digital Repository) provides the OCR technology (using Tesseract software) adapted to the requirements of herbarium specimen images and requires minimal installation in each institution. This is what we propose to make available to botanists with our portal. The goal for a museum is to be able to submit a large number of scanned images easily to a long-term archiving system in order to automatically obtain OCR texts and retrieve them by a full text search on an open data portal. Most of the images are provided for reuse through CC-BY licenses. In each case, the rights of reuse associated with the data are specified in associated metadata. This pilot was an opportunity to test the long-term storage service eTDR provided by CINES. The services (B2SAFE, B2Handle) developed by EUDAT were used to facilitate the transfer of data to the storage repository and to provide indexing services for access to that repository. This workflow that has been tested for the european project ICEDIG is presented as a poster: See the document (Suppl. material 1).

Boston Symphony Orchestra Archives (review)

IntroductionThe Boston Symphony Orchestra Ar - chives website currently contains digital scans of concert programs from 1881 to 2011, and over 2,000 photographic images. This online presentation, which uses OCLC's CONTENTdm platform,1 documents the activities of the Boston Sym phony Orchestra (BSO) as well as the Boston Sym - phony Chamber Players, the Boston Pops and Boston Pops Esplanade Orchestras, the Tanglewood Festival, and the Tanglewood Music Center Orchestra. Although the orchestra's physical archival collection also includes correspondence, press clippings, posters, musical instruments, and recordings, these formats are not represented online on this site at this time. However, the website does state that expansion of its digital collection is in process,2 and some of those materials are represented on the orchestra's main web site.3Embedded within this site is a separate performance history search interface called HENRY (named after the orchestra's founder, Henry Lee Higginson), which allows users to search information on performances from 21 October 1881, through the current season. Its database is updated on a monthly basis. The development of HENRY was funded by the National Endowment for the Humanities and the Alfred P. Sloan Foundation.NavigationIt is worth noting that the BSO Archives site's URL is http://collections.bso.org, but HENRY has a separate URL, http://archives .bso.org. These address names have the potential to confuse users.On the main page of the BSO Archives section (http://collections.bso.org), there are links to the two collections-Images and Program Books-with a brief description and a scrollable thumbnail image gallery for both. Above these are a search box with an option for Advanced Search, links to the About page, HENRY, and Browse All.Clicking on the respective collection link brings up a landing page that, like the main page, features a rotating gallery of thumbnails and a description of the collection's contents, and in addition presents a section of suggested topics to browse. On the right side of the page is a static grid of recently digitized items, and a handy RSS link so that users can be notified when new items are added to the collection.The Browse All link brings the user to a table of thumbnails, but its main function is as a search page for the collections, which will be discussed below. It is easier to browse and get a better sense of what is available through the suggested topics list.One suggested link on the Program Books page is the program for the first BSO concert, held on 22 October 1881.4 The item can be viewed as multiple PDF files with an Adobe Acrobat plug-in. There is also a tab to view as text, which is an unedited Optical Character Recognition (OCR) text file. Next to this tab is a text search box. As OCR text files often have some scanning errors, executing a word search within the program page can be hit or miss. The document scans, however, are of excellent quality, and can be zoomed in or out using Adobe's tools.Each page of the program is its own PDF file, and thumbnails of each file is in a vertical tray on the right side of the Web page. A Content tab is also located at the top of the tray, which simply supplies links to each PDF file as text, rather than as thumbnail images. The link names are not descriptive, however; each file is only known as seq. 1, seq. 2, and so forth.Options are available to view the program book in page-flip mode and with the PDF and text files presented side by side for comparison. There are also buttons for downloading the entire program book (as a multi-page PDF), and printing either the current page or the entire book. A page or book can be saved to a Favorites folder by clicking on Save Page or Save to Favorites, respectively. The Reference URL link generates permalinks and HTML to copy and paste, and clicking on Share pulls up functionality from the Add This tool, allowing the user to share the item through an array of social media. …

Loose, Falling Characters and Sentences: The Persistence of the OCR Problem in Digital Repository E-Books

The electronic conversion of scanned image files to readable text using optical character recognition (OCR) software and the subsequent migration of raw OCR text to e-book text file formats are key remediation or media conversion technologies used in digital repository e-book production. Despite real progress, the OCR problem of reliability and accuracy in OCR-derived e-book text and metadata persists. This paper examines a selection of digitized e-books in several prominent digital repositories and discusses the impact of OCR technology on e-book text file formats, metadata, and the online reading experience.

The use of Optical Character Recognition (OCR) in the digitisation of herbarium specimen labels.

At the Royal Botanic Garden Edinburgh (RBGE) the use of Optical Character Recognition (OCR) to aid the digitisation process has been investigated. This was tested using a herbarium specimen digitisation process with two stages of data entry. Records were initially batch-processed to add data extracted from the OCR text prior to being sorted based on Collector and/or Country. Using images of the specimens, a team of six digitisers then added data to the specimen records. To investigate whether the data from OCR aid the digitisation process, they completed a series of trials which compared the efficiency of data entry between sorted and unsorted batches of specimens. A survey was carried out to explore the opinion of the digitisation staff to the different sorting options. In total 7,200 specimens were processed.When compared to an unsorted, random set of specimens, those which were sorted based on data added from the OCR were quicker to digitise. Of the methods tested here, the most successful in terms of efficiency used a protocol which required entering data into a limited set of fields and where the records were filtered by Collector and Country. The survey and subsequent discussions with the digitisation staff highlighted their preference for working with sorted specimens, in which label layout, locations and handwriting are likely to be similar, and so a familiarity with the Collector or Country is rapidly established.