Scholarlink

A multi-script writer verification system poses a captivating research challenge in the fields of pattern recognition and computer vision. It entails the intricate task of authenticating the identities of writers who produce handwriting samples in various scripts. This scenario introduces complexities demanding robust techniques for accurate recognition across diverse scripts. In West Bengal, the native language is Bangla, with Hindi serving as the second language, while English finds use in official contexts. This study tackles the issue of block-level tri-script (Bangla, Hindi, and English) writer verification systems and presents promising results. The research demonstrates that a combination of handcrafted features outperforms automatically derived features, owing to limitations in the writers within the JUDVLP-TLWVdb dataset. Experimental results indicate that the SMO classifier outperforms other classifiers such as simple logistics and KNN. A novel dataset for writer verification systems using the tri-script approach is introduced, achieving a peak verification accuracy of 91.50% through a combination of Radon Transform, HOG, LBP, and LPQ features. The overall performance of the tri-script approach reaches 91.80%. Furthermore, this study employs the Vision Transformer (ViT) model for writer recognition, showcasing superior performance in comparison to ViT using tri-level block images of the page.

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

Evaluating the Tri-Script Writer Verification System Using a Handcrafted Features and Vision Transformer Learning Approach

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Jaya Paul,
  • Kalpita Dutta,
  • Anasua Sarkar,
  • Kaushik Roy,
  • Nibaran Das

摘要

A multi-script writer verification system poses a captivating research challenge in the fields of pattern recognition and computer vision. It entails the intricate task of authenticating the identities of writers who produce handwriting samples in various scripts. This scenario introduces complexities demanding robust techniques for accurate recognition across diverse scripts. In West Bengal, the native language is Bangla, with Hindi serving as the second language, while English finds use in official contexts. This study tackles the issue of block-level tri-script (Bangla, Hindi, and English) writer verification systems and presents promising results. The research demonstrates that a combination of handcrafted features outperforms automatically derived features, owing to limitations in the writers within the JUDVLP-TLWVdb dataset. Experimental results indicate that the SMO classifier outperforms other classifiers such as simple logistics and KNN. A novel dataset for writer verification systems using the tri-script approach is introduced, achieving a peak verification accuracy of 91.50% through a combination of Radon Transform, HOG, LBP, and LPQ features. The overall performance of the tri-script approach reaches 91.80%. Furthermore, this study employs the Vision Transformer (ViT) model for writer recognition, showcasing superior performance in comparison to ViT using tri-level block images of the page.