Navigating the Terrain: A Comprehensive Survey of Deep Learning for Image Forgery Detection

The increase in image forgery is due to the emergence of digital image processing technology, which necessitates the need to detect such fraudulent activities. This study proposes a new approach to address this challenge using Python and a Convolutional Neural Network (CNN) model. The proposed CNN model has achieved a training accuracy of 99% and a testing accuracy of 90%, demonstrating its effectiveness in recognizing the authenticity of the images. The data used in this study includes 12,615 images, including 7,492 real images (real) and 5,123 falsified (fake) images. To improve the accuracy of the proposed approach, error analysis is included as a preliminary step. All the input images are converted to standard 256x256 resolution and ELA is applied. ELA helps to identify the main areas in an image that exhibit different levels of compression. An undamped image shows good compression in all areas. Deviations from this consistency may indicate numerical manipulation. The resulting images are then stored as NumPy arrays for later analysis. The proposed scheme uses a combination of labels developed using CNN, deep learning and ELA. This combination allows the proposed model to not only achieve high accuracybut also provide an in-depth insight into specific areas of the image.