Scholarlink

<p>In optical character recognition (OCR), cursive recognition is a difficult area to perform since cursive writing contains a lot of fluidity and variability in its usage. This paper proposes a two-model architecture framework that uses hand-crafted properties in combination with deep learning to distinguish cursive forms in the English handwriting. The study evaluates the application of an ordinary Histogram of Oriented Gradients (HOG) model and VGG-19 convolutional neural net to test it with a cursive text dataset with a large range of stylistic differences. An extensive and diverse range of data that comprised of more than 1000 samples of handwriting were gathered and preprocessed to have a uniform input (into the model) by converting in grayscale, normalization and removal of noise. As the results show, the VGG-19 model was far more effective than the one based on HOG and, alongside that, demonstrated strong performance when it came to recognizing complex styles of writing, having an accuracy of 95.4% as opposed to 91.2%. This study will aid the development of the OCR domain and it has practical application implications related to document digitization, learning, and archival retention.</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

Improving Cursive Handwriting Recognition: A Dual Model Approach Using HOG Features and VGG-19

  • 期刊论文

  • 发表日期：

    2026年1月9日

查看全文
  • Chiguru Aparna,
  • K Rajchandar

摘要

In optical character recognition (OCR), cursive recognition is a difficult area to perform since cursive writing contains a lot of fluidity and variability in its usage. This paper proposes a two-model architecture framework that uses hand-crafted properties in combination with deep learning to distinguish cursive forms in the English handwriting. The study evaluates the application of an ordinary Histogram of Oriented Gradients (HOG) model and VGG-19 convolutional neural net to test it with a cursive text dataset with a large range of stylistic differences. An extensive and diverse range of data that comprised of more than 1000 samples of handwriting were gathered and preprocessed to have a uniform input (into the model) by converting in grayscale, normalization and removal of noise. As the results show, the VGG-19 model was far more effective than the one based on HOG and, alongside that, demonstrated strong performance when it came to recognizing complex styles of writing, having an accuracy of 95.4% as opposed to 91.2%. This study will aid the development of the OCR domain and it has practical application implications related to document digitization, learning, and archival retention.