DocCreator: A New Software for Creating Synthetic Ground-Truthed Document Images

Abstract

Most digital libraries that provide user-friendly interfaces, enabling quick and intuitive access to their resources, are based on Document Image Analysis and Recognition (DIAR) methods. Such DIAR methods need ground-truthed document images to be evaluated/compared and, in some cases, trained. Especially with the advent of deep learning-based approaches, the required size of annotated document datasets seems to be ever-growing. Manually annotating real documents has many drawbacks, which often leads to small reliably annotated datasets. In order to circumvent those drawbacks and enable the generation of massive ground-truthed data with high variability, we present DocCreator, a multi-platform and open-source software able to create many synthetic image documents with controlled ground truth. DocCreator has been used in various experiments, showing the interest of using such synthetic images to enrich the training stage of DIAR tools.

An Analysis of the Performance of Named Entity Recognition over OCRed Documents

“…More specifically, we extracted raw text from the test set and converted it into images. These images have been contaminated used the DocCreator tool 1 developed by Journet et al [7]. The tool allowed to add four common types of OCR degradation related to storage conditions or poor quality of printing materials that may be present on digital libraries: Character degradation adds small ink spots on characters in order to simulate degradation due to the age of the document or the use of an incorrectly set scanner.…”

Impact of OCR Quality on Named Entity Linking

“…We have used the DocCreator tool [12] to add four common types of OCR degradation that may be present on digital libraries material:…”

“…Each @ in the ground truth indicates the insertion of one character by the OCR while @ in the noisy text indicates that one character has been deleted from the original text. All data used in this work are available for public 12 . In order to evaluate the quality of the text generated by an OCR engine, one of the most popular measures is Character Error Rate (CER) [11].…”

Semi-Synthetic Data Augmentation of Scanned Historical Documents

“…However, in some cases, cursive fonts may not be available and the image structure can be hard to capture when there are a lot of variations in the dataset. Similarly, Journet et al [11] generate synthetic documents by extracting layout and characters using the Tesseract OCR. Then, the image is inpainted to recover the background image.…”