Fine – Grained Forgery Localization in Images Using CNN - SVM Approach

Nowadays, there is an unheard-of increase in the consumption of digital photos due to the widespread usage of gadgets like smartphones and tablets. Moreover, editing such content is now easier than ever thanks to the introduction of affordable, user-friendly picture alteration tools. A portion of these photos have been altered to the point where the human eye is unable to notice. Furthermore, social media platforms have simplified the process of distributing them to a wider audience. Therefore, it is crucial to create automated techniques that can identify these types of forgeries. In this work, we use Convolutional Neural Networks (CNN), a deep learning approach, to detect and pinpoint copy-move and splicing image forgeries. When using the CNN-based approach, an SVM classifier is given a feature representation from the network, which helps to determine if an image is real or fake. The tampered area is identified and returned if forgery is found. After a picture is determined to be forged, segmentation is applied to locate the spliced area. Then the area is masked and localized from the original image. This work has achieved an accuracy of 92.69%.