Transfer Learning for Efficient and Accurate Image Forgery Detection

Visuals are at the heart of information dissemination, especially on social platforms, but their credibility is not easy to ascertain because of their manipulability. Most of the forgery detection research done is mostly tailored to only a certain type of forgery and this limits its applicability in real life scenarios. The purpose of this study is to explore the use of transfer learning techniques of deep learning in improving the detection of forgeries. It is Iso used for finding the differences in compression quality in pre-trained models by generating features from them. The method tests eight models for binary classification, and MobileNetV2 has the most accurate performance of 95% because of its lightweight nature and efficiency. Results of the experiment indicate that the accuracy and computational efficiency of the proposed method is satisfactory and surpasses the performance of existing state of the art approaches and can therefore be used in development for environments and scenarios where resources are limited.