Scholarlink

<p>Copy Move Forgery (CMF) in one of the most widely used types of digital image manipulation, in which portions from the same image are duplicated to hide or recreate objects. This paper proposes an effective Copy-Move Forgery Detection (CMFD) approach using CLR optimized ResNeXt-50 convolutional neural network. The proposed method intends to achieve high detection accuracy while reducing model complexity and training time. Experiments were conducted on multiple benchmark datasets, including MICC-F220, MICC-F600, MICC-F2000, MICC combined, and CoMoFoD v2. The ResNeXt-50 model achieved a peak accuracy of 98.9% on CoMoFoD v2 within 8 epochs and a testing accuracy of 99.5%. For MICC-F220, the training and testing accuracies were 96.8% and 97.27%, respectively. On MICC-F600, the model achieved 98.1% training accuracy and 98.5% testing accuracy. For MICCF-2000, the training and testing accuracies were 98.2% and 98.4%, respectively, while the MICC combinational dataset yielded accuracies of 95.7% and 96.09%. These results underscore the potential of ResNeXt-50, combined with CLR, for efficient and accurate CMFD.</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 Deep learning approach for copy-move forgery detection using ResNeXt-50 and CLR scheduling

  • 期刊论文

  • 发表日期：

    2026年4月2日

查看全文
  • O. J. Sabu,
  • Thomas Monoth

摘要

Copy Move Forgery (CMF) in one of the most widely used types of digital image manipulation, in which portions from the same image are duplicated to hide or recreate objects. This paper proposes an effective Copy-Move Forgery Detection (CMFD) approach using CLR optimized ResNeXt-50 convolutional neural network. The proposed method intends to achieve high detection accuracy while reducing model complexity and training time. Experiments were conducted on multiple benchmark datasets, including MICC-F220, MICC-F600, MICC-F2000, MICC combined, and CoMoFoD v2. The ResNeXt-50 model achieved a peak accuracy of 98.9% on CoMoFoD v2 within 8 epochs and a testing accuracy of 99.5%. For MICC-F220, the training and testing accuracies were 96.8% and 97.27%, respectively. On MICC-F600, the model achieved 98.1% training accuracy and 98.5% testing accuracy. For MICCF-2000, the training and testing accuracies were 98.2% and 98.4%, respectively, while the MICC combinational dataset yielded accuracies of 95.7% and 96.09%. These results underscore the potential of ResNeXt-50, combined with CLR, for efficient and accurate CMFD.