Detection of Image Splicing Forgeries Based on Deep Learning with Edge Detector

Digital image forgery is on the rise, facilitated by the widespread use of software tools for image manipulation. Various software applications make image forgery increasingly accessible, with one commonly encountered technique being splicing. The rampant forgery of images on social media and online platforms threatens public trust in visual information. Forgery detection is crucial in restoring public confidence in images and visual content. Several methods, particularly those focusing on splicing techniques, involve digital analysis and forensic techniques for forgery detection. This research proposes a model for detecting digitally manipulated images, specifically those altered through splicing and authentic images. Our approach uses deep learning models and an edge detector, enabling it to learn forgery patterns and identify manipulated images. Combining the Convolutional Neural Networks (CNN) model and the Canny edge detector proves to be a robust approach for image forgery detection. The CNN model assists in comprehending complex patterns arising from splicing, while edge detection techniques provide information about suspicious boundary areas. To enhance the reliability of the model, we integrated two datasets, namely CASIAv1 and CASIAv2. Simulation results demonstrate that our CNN model achieves a testing accuracy of 84.92%, representing a significant 1.11% improvement compared to a CNN model without integrating the Canny edge detector. This research contributes to ongoing efforts to develop practical tools for detecting digital image forgery, thereby enhancing the reliability of visual information in the digital age.