Image Forgery Detection Using Integrated Convolution-LSTM (2D) and Convolution (2D)

Abstract

Digital forensics and computer vision must explore image forgery detection and their related technologies. Image fraud detection is expanding as sophisticated image editing software becomes more accessible. This makes changing photos easier than with the older methods. Convolution LSTM (1D) and Convolution LSTM (2D) + Convolution (2D) are popular deep learning models. We tested them using the public CASIA.2.0 image forgery database. ConvLSTM (2D) and its combination outperformed ConvLSTM (1D) in accuracy, precision, recall, and F1-score. We also provided a related work on image forgery detection models and methods. We also reviewed publicly available datasets used in picture forgery detection research, highlighting their merits and drawbacks. Our investigation revealed the state of picture fraud detection and the deep learning models that worked well. Our work greatly impacts fraudulent photo detection. First, it highlights how important deep learning models are for picture forgery detection. Second, ConvLSTM (2D) + Conv (2D) detect image forgeries better than ConvLSTM (1D). Finally, our dataset analysis and proposed integrated approach help research construct more effective and accurate picture forgery detection systems.