RDIFR: Robust Digital Image Forgery Recognition System for Image Splicing Using Deep Learning

Image forgery is the deceptive alteration of digital images, involving techniques like adding, removing, or modifying elements within the image. It is done for purposes such as creating misleading photos or falsifying evidence. Recognizing image forgery is essential to maintaining the authenticity and reliability of images in various fields. Image splicing is a type of digital image forgery in which different parts of multiple images are combined to create a new composite image. In this paper, we introduce a robust digital Image Forgery Recognition (RDIFR) system for image splicing recognition. The input image in our model passes through the error level analysis (ELA) module to apply image compression and then enters the feature extraction module, where we apply various deep learning-based models. The VGG-19 model achieved our best accuracy of 96.86% by applying L2 regularization and using a Nesterov-accelerated Adaptive Moment Estimation (NADAM) optimizer; however, it consumes 27 seconds per epoch, which is a high time consumption. Therefore, the CNN model with dropouts becomes our best choice. It has a slightly lower accuracy of 96.52 %, but it consumes only 8 seconds per epoch.