Detection of copy-move forgery in digital images using different computer vision approaches

Image forgery detection approaches are many and varied, but they generally all serve the same objectives: detect and localize the forgery. Copy-move forgery detection (CMFD) is widely spread and must challenge approach. In this thesis, We first investigate the problems and the challenges of the existed algorithms to detect copy-move forgery in digital images and then we propose integrating multiple forensic strategies to overcome these problems and increase the efficiency of detecting and localizing forgery based on the same image input source. Test and evaluate our copy-move forgery detector algorithm presented the outcome that has been enhanced by various computer vision field techniques. Because digital image forgery is a growing problem due to the increase in readily-available technology that makes the process relatively easy for forgers, we propose strategies and applications based on the PatchMatch algorithm and deep neural network learning (DNN). We further focus on the convolutional neural network (CNN) architecture approach in a generative adversarial network (GAN) and transfer learning environment. The F-measure score (FM), recall, precision, accuracy, and efficiency are calculated in the proposed algorithms and compared with a selection of literature algorithms using the same evaluation function in order to make a fair evaluation. The FM score achieves 0.98, with an efficiency rate exceeding 90.5% in most cases of active and passive forgery detection tasks, indicating that the proposed methods are highly robust. The output results show the high efficiency of detecting and localizing the forgery across different image formats for active and passive forgery detection. Therefore, the proposed methods in this research successfully overcome the main investigated issues in copy-move forgery detection as such: First, increase efficiency in copy-move forgery detection under a wide range of manipulation process to a copy-moved image. Second, detect and localized the copy-move forgery patches versus the pristine patches in the forged image. Finally, our experiments show the overall validation accuracy based on the proposed deep learning approach is 90%, according to the iteration limit. Further enhancement of the deep learning and learning transfer approach is recommended for future work