Scholarlink

<p>The ability to detect text in images or video is useful in a wide range of applications since the learned properties of deep learning can manage to point out the textual object. However, many existing techniques have mediocre performance when applied to recognise arbitrary-shaped text in pictures. This is mostly constrained by their textual representations which are horizontal boxes, tilted rectangle and quadrilaterals. It presents Deep-Learned Fusion Attention Network (DLFANet) to recognize the distinguishing features of arbitrary-shaped text, through a small architecture named shared network which is further fine-tuned by the suggested Feature Attention Module Enhancement (FAME). In addition, the Final Feature Module (FFM) also includes Attention Detection Head (ADH) and Geometry Aware Pixel Network (GAPN) to properly recognize the location of the text. The testing of the proposed method on the include benchmark Total-Text, CTW 1500 and ICDAR 2015 data sets presents better results when compared to other existing advanced algorithms. </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

DLFANet: a Deep-Learned Fusion Attention Network for Robust Text Detection in Complex Scenes

  • 期刊论文

  • 发表日期：

    2026年1月21日

查看全文
  • Ramalakshmi Subbukalai,
  • Vani Vijayan

摘要

The ability to detect text in images or video is useful in a wide range of applications since the learned properties of deep learning can manage to point out the textual object. However, many existing techniques have mediocre performance when applied to recognise arbitrary-shaped text in pictures. This is mostly constrained by their textual representations which are horizontal boxes, tilted rectangle and quadrilaterals. It presents Deep-Learned Fusion Attention Network (DLFANet) to recognize the distinguishing features of arbitrary-shaped text, through a small architecture named shared network which is further fine-tuned by the suggested Feature Attention Module Enhancement (FAME). In addition, the Final Feature Module (FFM) also includes Attention Detection Head (ADH) and Geometry Aware Pixel Network (GAPN) to properly recognize the location of the text. The testing of the proposed method on the include benchmark Total-Text, CTW 1500 and ICDAR 2015 data sets presents better results when compared to other existing advanced algorithms.