Scholarlink

<p>With the growing need for accurate digital text retrieval from images, OCR (optical character recognition) plays a critical role in fields such as document digitization, automated data processing, and information retrieval. The given paper introduces optical OCR system, comprising Convolutional Recurrent Neural Network (CRNN) embedded with a Word Beam Search (WBS) decoder for extraction of text. The framework performs preprocessing, feature extraction, and post-processing strategies that enhance the accuracy and robustness of text extraction. Preprocessing techniques and post-processing adjustments significantly improved the model's ability to handle diverse image formats, including scanned documents and screenshots. The proposed system demonstrates an overall accuracy of 92%, with macro and weighted averages across precision, recall, and F1-scores also at 92%. Individual class evaluations reveal notable performance consistency, with precision ranging from 0.79 to 0.98, recall between 0.79 and 0.96, and F1-scores from 0.79 to 0.94 across 26 alphabetic characters. This research highlights the system's potential for real-world applications, demonstrating its scalability, precision, and efficiency in extracting text data across various formats, ensuring high OCR accuracy in automated data processing.</p>

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

OCR-CRNN (WBS): an optical character recognition system based on convolutional recurrent neural network embedded with word beam search decoder for extraction of text

  • 期刊论文

  • 发表日期：

    2025年5月8日

查看全文
  • Mahesh Kumar,
  • Shalu,
  • Ajay Dureja,
  • Rachna Narula,
  • Shyla,
  • Ravi Arora

摘要

With the growing need for accurate digital text retrieval from images, OCR (optical character recognition) plays a critical role in fields such as document digitization, automated data processing, and information retrieval. The given paper introduces optical OCR system, comprising Convolutional Recurrent Neural Network (CRNN) embedded with a Word Beam Search (WBS) decoder for extraction of text. The framework performs preprocessing, feature extraction, and post-processing strategies that enhance the accuracy and robustness of text extraction. Preprocessing techniques and post-processing adjustments significantly improved the model's ability to handle diverse image formats, including scanned documents and screenshots. The proposed system demonstrates an overall accuracy of 92%, with macro and weighted averages across precision, recall, and F1-scores also at 92%. Individual class evaluations reveal notable performance consistency, with precision ranging from 0.79 to 0.98, recall between 0.79 and 0.96, and F1-scores from 0.79 to 0.94 across 26 alphabetic characters. This research highlights the system's potential for real-world applications, demonstrating its scalability, precision, and efficiency in extracting text data across various formats, ensuring high OCR accuracy in automated data processing.