Scholarlink

<p>This paper presents a robust system for recognizing handwritten numerals by leveraging advanced feature extraction techniques. The recognition process begins with generating a skeleton representation of the numeral, facilitating the extraction of essential and meaningful features. A hybrid approach is employed, integrating four distinct feature types: centroid features, diagonal features, zoning features, and peak extent-based features. This fusion enhances the system’s ability to distinguish numerals accurately. For classification, a Support Vector Machine (SVM) classifier is utilized, as it is well-suited for high-dimensional feature spaces. The system’s effectiveness was evaluated on a dataset containing 7,500 samples of isolated handwritten numerals. Experimental results demonstrate the proposed system’s high accuracy, achieving a recognition rate of 97.46% using a fivefold cross-validation technique. This approach offers a promising solution for applications requiring efficient and accurate handwritten numeral recognition. Additionally, this study introduces a hybrid feature extraction approach that combines statistical and deep learning techniques to improve the quality of reconstructed signals, particularly in the presence of Gaussian and impulse noise. The methodology was tested on both grayscale and color images, demonstrating significant improvements over traditional methods.</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

Handwritten Numeral Recognition System Using SVM with Multi-Feature Extraction

  • 期刊论文

  • 发表日期：

    2025年6月29日

查看全文
  • Kaifeng Chen,
  • Ahmed Alkhayyat,
  • Aman Shankhyan,
  • Prabhat Kumar Sahu,
  • V. Vivek,
  • Amrita Singh,
  • Protyay Dey,
  • B. Bharathi

摘要

This paper presents a robust system for recognizing handwritten numerals by leveraging advanced feature extraction techniques. The recognition process begins with generating a skeleton representation of the numeral, facilitating the extraction of essential and meaningful features. A hybrid approach is employed, integrating four distinct feature types: centroid features, diagonal features, zoning features, and peak extent-based features. This fusion enhances the system’s ability to distinguish numerals accurately. For classification, a Support Vector Machine (SVM) classifier is utilized, as it is well-suited for high-dimensional feature spaces. The system’s effectiveness was evaluated on a dataset containing 7,500 samples of isolated handwritten numerals. Experimental results demonstrate the proposed system’s high accuracy, achieving a recognition rate of 97.46% using a fivefold cross-validation technique. This approach offers a promising solution for applications requiring efficient and accurate handwritten numeral recognition. Additionally, this study introduces a hybrid feature extraction approach that combines statistical and deep learning techniques to improve the quality of reconstructed signals, particularly in the presence of Gaussian and impulse noise. The methodology was tested on both grayscale and color images, demonstrating significant improvements over traditional methods.