Image Splicing and Copy-Move Image Forgery Identification Methods based on Deep Learning: A Survey


  • Rupali Bora K K Wagh Insitute of Engineering
  • Mahesh Sanghavi SNJB, Chandwad


Copy-move, Image Splicing, Convolutional Neural Network (CNN), Long Short Term Memory (LSTM), Region-based Convolutional Neural Network (R-CNN)


With the advancements of technology in current era, everyone faces a challenge to identify digitally manipulated images. It is not easy to discriminate the original and forged images. For digital image tampering, image splicing and copy-move forgeries are very much well-known and common techniques. Image forgery is detected and spotted based on feature descriptor of an image. It is a concise and important local descriptor which is to be applied to grasp hierarchical representations from the input images. The significant correlation among nearby pixels has been identified by deep learning-based methods. It prefers locally grouped networks rather than one-to-one networks among all pixels. A convolution operation can be implemented by sharing weights to produce the output feature map. Many well-known deep learning-based backbone architectures, such as CNN, R-CNN, LSTM, U-Net, encoder-decoder have been implemented in the past for detection and localization of these types of image forgeries. To detect copy-move and splicing forgeries can be proposed based on these existing frameworks. The various performance parameters like accuracy, F1-score, AUC will be computed and compared. Many available standard forged image datasets comprising of images of these forgery types, viz. CASIA_v2, CoMoFoD, Columbia, IMD, NIST16 will be used for training purpose. Different training-testing datasets will be used for model working to check effectiveness of model.