ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper 다중 언어로 학습된 이미지 기반 문자인식 모델의 성능 분석
Cited - time in scopus Download 22 time Share share facebook twitter linkedin kakaostory
Authors
오현호, 윤준석, 배유석, 김형일
Issue Date
2023-06
Citation
대한전자공학회 학술 대회 (하계) 2023, pp.1058-1061
Language
Korean
Type
Conference Paper
Project Code
23HS1600, Development of High Performance Visual Discovery Platform for Realtime and Large-Scale Data Analysis and Prediction, Bae Yu Seok
Abstract
As optical character recognition technology is one of the pattern recognition technologies that have been studied for a long time, deep learning-based scene text recognition (STR) has obtained much attention due to the advances in deep learning and the availability of large-scale text image datasets. However, as the previous deep learning-based STR algorithms are trained with a single language (e.g., English or Chinese), it is limited to applying to a real-world environment where multi-languages appear simultaneously. In this paper, we retrain the off-the-shelf STR deep networks with multilanguages (specifically Korean and English) for the purpose of real-world applications such as text understanding for outdoor signs or banners. Then, the text classification performance of the retrained STR models and the effect of multilingual training are analyzed.
KSP Keywords
Classification Performance, Image datasets, Learning-based, Long Time, Off-the-shelf, Optical character recognition, Pattern recognition, Real-world applications, Scene text recognition, Text understanding, deep learning(DL)