Thai Scene Text Recognition with Character Combination

Chun Li, Hongjian Zhan, Kun Zhao, Yue Lu

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

2 Scopus citations

Abstract

In recent years, scene text recognition(STR) that recognizing character sequences in natural images is in great demand beyond various fields. However, most STR studies only focus on popular scripts like Chinese or English, too little attention has been paid to minority languages. In this paper, we address problems on Thai STR, and introduce a novel strategy called Thai Character Combination(TCC), which explore original characteristics of Thai text. Unlike most other scripts, characters in Thai text can be written both horizontally and vertically, which brings big challenges to current sequence-based text recognition methods. In order to reduce complexity of structure and alleviate the misalignment problem in attention-based methods, TCC intends to combine Thai characters that stack vertically to independent combined characters. Furthermore, we establish a Thai Scene Text(TST) dataset that collected from multiple scenarios to evaluate the performance of our proposed character modeling strategy. We conduct abundant experiments and analyses to compare the recognition performance of models with and without TCC. The results indicate the effectiveness of the proposed method from multiple perspectives, especially, TCC benefits a lot for long text recognition, and there is a substantial improvement in the recognition accuracy of entire string-level.

Original languageEnglish
Title of host publicationPattern Recognition and Computer Vision - 5th Chinese Conference, PRCV 2022, Proceedings
EditorsShiqi Yu, Jianguo Zhang, Zhaoxiang Zhang, Tieniu Tan, Pong C. Yuen, Yike Guo, Junwei Han, Jianhuang Lai
PublisherSpringer Science and Business Media Deutschland GmbH
Pages320-333
Number of pages14
ISBN (Print)9783031189128
DOIs
StatePublished - 2022
Event5th Chinese Conference on Pattern Recognition and Computer Vision, PRCV 2022 - Shenzhen, China
Duration: 4 Nov 20227 Nov 2022

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume13536 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference5th Chinese Conference on Pattern Recognition and Computer Vision, PRCV 2022
Country/TerritoryChina
CityShenzhen
Period4/11/227/11/22

Keywords

  • Scene text recognition
  • Thai Character Combination
  • Thai scene text dataset

Fingerprint

Dive into the research topics of 'Thai Scene Text Recognition with Character Combination'. Together they form a unique fingerprint.

Cite this