Augmenting Historical Alphabet Datasets Using Generative Adversarial Networks

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F60460709%3A41110%2F23%3A92542" target="_blank" >RIV/60460709:41110/23:92542 - isvavai.cz</a>

Result on the web

<a href="https://link.springer.com/chapter/10.1007/978-3-031-21438-7_11" target="_blank" >https://link.springer.com/chapter/10.1007/978-3-031-21438-7_11</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-031-21438-7_11" target="_blank" >10.1007/978-3-031-21438-7_11</a>

Result language

angličtina

Original language name

Augmenting Historical Alphabet Datasets Using Generative Adversarial Networks

Original language description

In this paper, we present a method for expanding small classification datasets. Every research project is based on data and methods, including text analysis. When analyzing historical texts in different alphabets, there are not always Optical Character Recognition algorithms available and, in many cases, such texts need to be transliterated and translated manually, or alternatively, an OCR algorithm can be developed. In order to create such an algorithm, a large volume of input data is needed - each alphabet consists of elementary data - either letters, vowels, or in some cases ideograms. The texts need to be segmented into such elements, and then, the elements are classified. In many cases, it is very difficult and time-costly to get a sufficient amount of data, and it is advisable to use augmentation methods. In our research, we propose using Generative Adversarial Network to expand a relatively small dataset of Palmyrene letters and prove that even by adding generated data equal to the third of size of the original dataset, the classification results are improved by 120 percent.

Czech name

—

Czech description

—

Type

D - Article in proceedings

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

—

Continuities

S - Specificky vyzkum na vysokych skolach

Publication year

2023

Confidentiality

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

Article name in the collection

Data Science and Algorithms in Systems

ISBN

978-3-031-21438-7

ISSN

2367-3389

e-ISSN

—

Number of pages

10

Pages from-to

132-141

Publisher name

Springer

Place of publication

Gewerbestrasse 11, 6330 Cham, Switzerland

Event location

online (Praha)

Event date

Jan 1, 2022

Type of event by nationality

WRD - Celosvětová akce

UT code for WoS article

—