Scholarlink

Image manipulation has brought up great challenges regarding the genuineness of digital media and forensic examination. This paper considers a holistic approach for the detection of an image forgery by using deep learning models, accompanied by preprocessing techniques encompassing Error Level Analysis (ELA). The proposed system employs CNN along with a hybrid CNN-LSTM model to classify images as authentic or tampered. ELA preprocessing enhances the visibility of compression artifacts, which tends to make detection easier. Using the CASIA v2 dataset, the CNN model has 94% testing accuracy, while the hybrid CNN-LSTM model has 88% accuracy. The current research highlights the efficiency with the integration of ELA with deep models regarding the detection of forgery and outlines the limitations and future courses of action in order to improve accuracy and generalization over different forgery techniques. The contribution of the work is to digital image forensics and the wider field of media integrity.

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

Image Forgery Detection Using ELA with CNN and CNN-LSTM Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Chalavadi Vatsalya Gayatri,
  • M. Gargi,
  • Cherukuri Kiran kumar,
  • Chittela Srinu

摘要

Image manipulation has brought up great challenges regarding the genuineness of digital media and forensic examination. This paper considers a holistic approach for the detection of an image forgery by using deep learning models, accompanied by preprocessing techniques encompassing Error Level Analysis (ELA). The proposed system employs CNN along with a hybrid CNN-LSTM model to classify images as authentic or tampered. ELA preprocessing enhances the visibility of compression artifacts, which tends to make detection easier. Using the CASIA v2 dataset, the CNN model has 94% testing accuracy, while the hybrid CNN-LSTM model has 88% accuracy. The current research highlights the efficiency with the integration of ELA with deep models regarding the detection of forgery and outlines the limitations and future courses of action in order to improve accuracy and generalization over different forgery techniques. The contribution of the work is to digital image forensics and the wider field of media integrity.