Scholarlink

<p>Ancient stone inscriptions are vital sources of traditional knowledge and historical heritage. However, because of the unrestricted stroke writings, age-related deterioration, and variability, their digital recognition is still difficult. Modern OCR technologies are inadequate for the accurate recognition of Tamil inscriptions. A Dynamic Profiling Bound (DPB) approach is proposed for character localization, achieving a high localization rate of 98.25% across varied inscription image conditions, including noise, deterioration, and compact writing. Using these localized characters, SignaryNet, a fine-tuned deep convolutional neural network (CNN), is developed to recognize 100 classes of ancient Tamil and Grantha characters based on Unicode mappings. SignaryNet outperforms the baseline Classic-CNN, achieving 98.61% validation and 96.81% prediction accuracy versus C-CNN’s 83.33% and 84.74%. The method proves effective on benchmark datasets and supports the digital preservation of heritage inscriptions, such as those on the Brihadisvara temple walls, facilitating their storage in accessible digital archives for historical research and conservation.</p><p></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

Leveraging digital acquisition and DPB based SignaryNet for localization and recognition of heritage inscription palaeography

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月23日

查看全文
  • S. Ezhilarasi,
  • P. Uma Maheswari

摘要

Ancient stone inscriptions are vital sources of traditional knowledge and historical heritage. However, because of the unrestricted stroke writings, age-related deterioration, and variability, their digital recognition is still difficult. Modern OCR technologies are inadequate for the accurate recognition of Tamil inscriptions. A Dynamic Profiling Bound (DPB) approach is proposed for character localization, achieving a high localization rate of 98.25% across varied inscription image conditions, including noise, deterioration, and compact writing. Using these localized characters, SignaryNet, a fine-tuned deep convolutional neural network (CNN), is developed to recognize 100 classes of ancient Tamil and Grantha characters based on Unicode mappings. SignaryNet outperforms the baseline Classic-CNN, achieving 98.61% validation and 96.81% prediction accuracy versus C-CNN’s 83.33% and 84.74%. The method proves effective on benchmark datasets and supports the digital preservation of heritage inscriptions, such as those on the Brihadisvara temple walls, facilitating their storage in accessible digital archives for historical research and conservation.