Scholarlink

<p>Coverless steganography seeks to conceal information without altering host media by exploiting inherent visual, textual, or audiovisual content. A promising direction encodes data through the detection and spatial arrangement of objects within images. However, the reliability of this approach depends strongly on object detection performance, which can be influenced by visual clutter, occlusion, and low image quality. This study investigates two advanced object detection and scene interpretation models-DEtection TRansformer (DETR) and Graph Region-based Convolutional Neural Network (Graph R-CNN)–for their potential applicability in future coverless steganography systems. Each model is evaluated independently for its ability to extract meaningful visual elements and relational cues for secure, content-based data representation. Experimental results indicate that DETR demonstrates 15.49% higher image-text semantic similarity than Graph R-CNN. While data embedding and retrieval are not yet implemented, the findings establish a proof of concept for integrating semantic scene understanding into next-generation coverless steganography frameworks, consistent with our pending patent.</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

A Novel Sequencing Framework for Multi-object Scene Interpretation in Coverless Image Steganography Using DETR and Graph R-CNN

  • 期刊论文

  • 发表日期：

    2026年1月13日

查看全文
  • H. Rama Moorthy,
  • Shreenath Acharya,
  • Kavana Shetty,
  • Jathin Salian,
  • Royston Cardoza,
  • N S Krishnaraj Rao,
  • Amrithkala M. Shetty

摘要

Coverless steganography seeks to conceal information without altering host media by exploiting inherent visual, textual, or audiovisual content. A promising direction encodes data through the detection and spatial arrangement of objects within images. However, the reliability of this approach depends strongly on object detection performance, which can be influenced by visual clutter, occlusion, and low image quality. This study investigates two advanced object detection and scene interpretation models-DEtection TRansformer (DETR) and Graph Region-based Convolutional Neural Network (Graph R-CNN)–for their potential applicability in future coverless steganography systems. Each model is evaluated independently for its ability to extract meaningful visual elements and relational cues for secure, content-based data representation. Experimental results indicate that DETR demonstrates 15.49% higher image-text semantic similarity than Graph R-CNN. While data embedding and retrieval are not yet implemented, the findings establish a proof of concept for integrating semantic scene understanding into next-generation coverless steganography frameworks, consistent with our pending patent.