Scholarlink

<p>This study proposes a Writer Identification System (WIS) specifically designed for handwritten Gurmukhi script, utilizing a hybrid set of handcrafted features and machine learning classifiers to support forensic and biometric authentication. A curated dataset comprising 21,000 samples of 35 fundamental Gurmukhi characters—each repeated five times by 60 distinct writers—forms the basis of this work. Four complementary feature extraction techniques—Zoning, Transition, Diagonal, and Peak Extent—are employed to encode spatial, directional, and structural characteristics of handwriting. These features are individually and jointly evaluated using three classifiers: Multi-Layer Perceptron (MLP), Artificial Neural Network (ANN), and Random Forest (RF). Among them, RF demonstrates superior performance, achieving a maximum accuracy of 94.39%, a True Positive Rate (TPR) of 94.26%, and a low False Positive Rate (FPR) of 0.38% when all features are combined. The experimental results affirm the effectiveness of multi-feature fusion and ensemble learning in enhancing writer identification accuracy. This work not only advances computational approaches for Gurmukhi script analysis but also lays a foundation for future extensions in behavioral biometrics, such as writer gender, handedness, and emotional state estimation.</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

Writer Identification from Handwritten Gurmukhi Script: A Machine Learning Approach

  • 期刊论文

  • 发表日期：

    2026年4月17日

查看全文
  • Girish Paliwal,
  • Kanta Prasad Sharma,
  • Ibrahim Oteir,
  • Mohd Shukri Ab Yajid,
  • Shivakrishna Dasi,
  • Ankur Srivastava,
  • Sankara Rao Palla,
  • Aman Shankhyan

摘要

This study proposes a Writer Identification System (WIS) specifically designed for handwritten Gurmukhi script, utilizing a hybrid set of handcrafted features and machine learning classifiers to support forensic and biometric authentication. A curated dataset comprising 21,000 samples of 35 fundamental Gurmukhi characters—each repeated five times by 60 distinct writers—forms the basis of this work. Four complementary feature extraction techniques—Zoning, Transition, Diagonal, and Peak Extent—are employed to encode spatial, directional, and structural characteristics of handwriting. These features are individually and jointly evaluated using three classifiers: Multi-Layer Perceptron (MLP), Artificial Neural Network (ANN), and Random Forest (RF). Among them, RF demonstrates superior performance, achieving a maximum accuracy of 94.39%, a True Positive Rate (TPR) of 94.26%, and a low False Positive Rate (FPR) of 0.38% when all features are combined. The experimental results affirm the effectiveness of multi-feature fusion and ensemble learning in enhancing writer identification accuracy. This work not only advances computational approaches for Gurmukhi script analysis but also lays a foundation for future extensions in behavioral biometrics, such as writer gender, handedness, and emotional state estimation.