Semantic text segmentation from synthetic images of full-text documents

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F49777513%3A23520%2F19%3A43958267" target="_blank" >RIV/49777513:23520/19:43958267 - isvavai.cz</a>

Výsledek na webu

<a href="http://proceedings.spiiras.nw.ru/index.php/sp/article/view/4527/2627" target="_blank" >http://proceedings.spiiras.nw.ru/index.php/sp/article/view/4527/2627</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.15622/sp.2019.18.6.1381-1406" target="_blank" >10.15622/sp.2019.18.6.1381-1406</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Semantic text segmentation from synthetic images of full-text documents

Popis výsledku v původním jazyce

An algorithm (divided into multiple modules) for generating images of fulltext documents is presented. These images can be used to train, test, and evaluate models for Optical Character Recognition (OCR). The algorithm is modular, individual parts can be changed and tweaked to generate desired images. A method for obtaining background images of paper from already digitized documents is described. For this, a novel approach based on Variational AutoEncoder (VAE) to train a generative model was used. These backgrounds enable the generation of similar background images as the training ones on the fly. The module for printing the text uses large text corpora, a font, and suitable positional and brightness character noise to obtain believable results (for natural-looking aged documents). A few types of layouts of the page are supported. The system generates a detailed, structured annotation of the synthesized image. Tesseract OCR to compare the real-world images to generated images is used. The recognition rate is very similar, indicating the proper appearance of the synthetic images. Moreover, the errors which were made by the OCR system in both cases are very similar. From the generated images, fully-convolutional encoder-decoder neural network architecture for semantic segmentation of individual characters was trained. With this architecture, the recognition accuracy of 99.28% on a test set of synthetic documents is reached.

Název v anglickém jazyce

Semantic text segmentation from synthetic images of full-text documents

Popis výsledku anglicky

An algorithm (divided into multiple modules) for generating images of fulltext documents is presented. These images can be used to train, test, and evaluate models for Optical Character Recognition (OCR). The algorithm is modular, individual parts can be changed and tweaked to generate desired images. A method for obtaining background images of paper from already digitized documents is described. For this, a novel approach based on Variational AutoEncoder (VAE) to train a generative model was used. These backgrounds enable the generation of similar background images as the training ones on the fly. The module for printing the text uses large text corpora, a font, and suitable positional and brightness character noise to obtain believable results (for natural-looking aged documents). A few types of layouts of the page are supported. The system generates a detailed, structured annotation of the synthesized image. Tesseract OCR to compare the real-world images to generated images is used. The recognition rate is very similar, indicating the proper appearance of the synthetic images. Moreover, the errors which were made by the OCR system in both cases are very similar. From the generated images, fully-convolutional encoder-decoder neural network architecture for semantic segmentation of individual characters was trained. With this architecture, the recognition accuracy of 99.28% on a test set of synthetic documents is reached.

Druh

J_SC - Článek v periodiku v databázi SCOPUS

CEP obor

—

OECD FORD obor

20205 - Automation and control systems

Projekt

Výsledek vznikl pri realizaci vícero projektů. Více informací v záložce Projekty.

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)<br>S - Specificky vyzkum na vysokych skolach<br>I - Institucionalni podpora na dlouhodoby koncepcni rozvoj vyzkumne organizace

Rok uplatnění

2019

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název periodika

SPIIRAS Proceedings

ISSN

2078-9181

e-ISSN

—

Svazek periodika

18

Číslo periodika v rámci svazku

6

Stát vydavatele periodika

RU - Ruská federace

Počet stran výsledku

26

Strana od-do

1380-1405

Kód UT WoS článku

—

EID výsledku v databázi Scopus

2-s2.0-85078454715