Scholarlink

By integrating a text detection model with text recognition models, the TH-DL text recognition system framework was designed and implemented for different tasks. Different text recognition models proposed in the book are compared. The experimental results show that PREN2D has achieved the highest recognition accuracy for word-level scene text images, whereas ARN-DTRN has achieved the best performance for sentence-level text images. A multilingual scene text recognition system based on the TH-DL framework was ranked first on the RRC-MLT-2019 leaderboard for the end-to-end text detection and recognition task. An Arabic video subtitle recognition system based on the TH-DL framework ranked first in the ICDAR 2017 and ICPR 2020 competitions on Arabic text detection and recognition in multiresolution video frames.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

TH-DL Multilingual Text Recognition System Framework

  • 章节

  • 发表日期：

    2025年

查看全文
  • Liangrui Peng,
  • Ruijie Yan

摘要

By integrating a text detection model with text recognition models, the TH-DL text recognition system framework was designed and implemented for different tasks. Different text recognition models proposed in the book are compared. The experimental results show that PREN2D has achieved the highest recognition accuracy for word-level scene text images, whereas ARN-DTRN has achieved the best performance for sentence-level text images. A multilingual scene text recognition system based on the TH-DL framework was ranked first on the RRC-MLT-2019 leaderboard for the end-to-end text detection and recognition task. An Arabic video subtitle recognition system based on the TH-DL framework ranked first in the ICDAR 2017 and ICPR 2020 competitions on Arabic text detection and recognition in multiresolution video frames.