Scholarlink

<p>In spite of advancements in the last decade regarding character recognition, there is still relatively little research about Handwritten Odia Script Recognition. Handwritten Odia Script has several factors that make it challenging to recognize such as curvilinear characters, high visual similarity between symbols and variability in individual handwriting style. Traditionally recognition systems have been based on Hand Crafted Features that are difficult to scale and require considerable Script Specific Expertise. In contrast, Deep Learning Methods can automatically learn Discriminative Features which allows for greater reliability and adaptability. This paper proposes an improved approach to character recognition using a Mish-Enhanced Xception (ME-Xception) Model which incorporates the Mish Activation Function within the Xception Framework to improve both Accuracy and Generalization. For Training and Testing the models, this work utilized benchmark datasets from ISI Kolkata, IIT Bhubaneswar, NIT Rourkela, and IIIT Bhubaneswar. Preprocessing was performed prior to training to remove Noise, and Srgan (Super-Resolution Generative Adversarial Network), was applied to Upscale Low Resolution Images from <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(64 \times 64\)</EquationSource> </InlineEquation> to <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(256 \times 256\)</EquationSource> </InlineEquation> Pixels. The resulting higher resolution images were provided to the ME-Xception Models. The experimental results show that the proposed technique consistently outperformed previously existing techniques for all datasets, and therefore demonstrates its suitability for Handwritten Character Recognition of Odia.</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 Odia handwritten character recognition with super-resolution GAN and mish-enhanced Xception network

  • 期刊论文

  • 发表日期：

    2025年11月23日

查看全文
  • Pragnya Ranjan Dash,
  • Rakesh Chandra Balabantaray

摘要

In spite of advancements in the last decade regarding character recognition, there is still relatively little research about Handwritten Odia Script Recognition. Handwritten Odia Script has several factors that make it challenging to recognize such as curvilinear characters, high visual similarity between symbols and variability in individual handwriting style. Traditionally recognition systems have been based on Hand Crafted Features that are difficult to scale and require considerable Script Specific Expertise. In contrast, Deep Learning Methods can automatically learn Discriminative Features which allows for greater reliability and adaptability. This paper proposes an improved approach to character recognition using a Mish-Enhanced Xception (ME-Xception) Model which incorporates the Mish Activation Function within the Xception Framework to improve both Accuracy and Generalization. For Training and Testing the models, this work utilized benchmark datasets from ISI Kolkata, IIT Bhubaneswar, NIT Rourkela, and IIIT Bhubaneswar. Preprocessing was performed prior to training to remove Noise, and Srgan (Super-Resolution Generative Adversarial Network), was applied to Upscale Low Resolution Images from \(64 \times 64\) to \(256 \times 256\) Pixels. The resulting higher resolution images were provided to the ME-Xception Models. The experimental results show that the proposed technique consistently outperformed previously existing techniques for all datasets, and therefore demonstrates its suitability for Handwritten Character Recognition of Odia.