Scholarlink

<p>Photographs are used as digital evidence in cybercrime investigations. Copy-move forging is an image tampering technique that involves hiding unwanted portions or reproducing desirable elements within the same image to create manipulated images that have been edited. Existing forgery detection techniques are less effective in detecting manipulated areas due to the large dimensions and low contrast of the images. This paper proposes a deep learning technique that provides an effective method for automatically detecting, localizing, and quantifying the extent of copy-move forgery in images. The proposed method utilizes an EfficientNet B0 architecture for feature extraction and a Feature Pyramidal Network to extract semantically rich features from the low-resolution features, thereby enabling the detection of the exact altered area in the image. The CoMoFoD and CASIA v2.0 datasets are used to conduct and test the planned work's effectiveness. The proposed method's performance is also analyzed using the Buster Net. The results reveal that the EfficientNet with FPN model outperforms the Buster Net model in detecting forgery in photos, achieving a sensitivity of 96.9 percent, whereas the sensitivity of the Buster Net model is 74.2 percent.</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

EfficientNet-FPN: a robust approach for detecting and locating copy-move forgery in digital images

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Sabeena M,
  • Lizy Abraham,
  • Syama S. R

摘要

Photographs are used as digital evidence in cybercrime investigations. Copy-move forging is an image tampering technique that involves hiding unwanted portions or reproducing desirable elements within the same image to create manipulated images that have been edited. Existing forgery detection techniques are less effective in detecting manipulated areas due to the large dimensions and low contrast of the images. This paper proposes a deep learning technique that provides an effective method for automatically detecting, localizing, and quantifying the extent of copy-move forgery in images. The proposed method utilizes an EfficientNet B0 architecture for feature extraction and a Feature Pyramidal Network to extract semantically rich features from the low-resolution features, thereby enabling the detection of the exact altered area in the image. The CoMoFoD and CASIA v2.0 datasets are used to conduct and test the planned work's effectiveness. The proposed method's performance is also analyzed using the Buster Net. The results reveal that the EfficientNet with FPN model outperforms the Buster Net model in detecting forgery in photos, achieving a sensitivity of 96.9 percent, whereas the sensitivity of the Buster Net model is 74.2 percent.