IMAGE FORGERY DETECTION USING DEEP LEARNING

Abstract

In the contemporary era, digital images constitute a primary means of disseminating information on social media platforms.However, this prevalence has given rise to a pressing issue-malicious software adept at fabricating images to propagate false information.Addressing this concern, existing literature has explored various digital image forgery detection techniques.Yet, many of these methods are confined to detecting singular types of forgery, such as image splicing or copy-move, which may not accurately reflect real-world scenarios.This paper introduces a novel approach to bolster digital image forgery detection by leveraging deep learning techniques through transfer learning.The goal is to simultaneously uncover two distinct types of image forgery.The proposed method hinges on identifying variations in the compressed quality of forged areas, typically deviating from the compressed quality of the rest of the image.A deep learning-based model is presented for forgery detection in digital images.This involves calculating the disparity between the original image and its compressed version to generate a featured image, serving as input to a pre-trained model.The pre-trained model undergoes training with its classifier removed, and a new fine-tuned classifier is introduced.A comparative analysis is conducted among eight different pre-trained models tailored for binary classification.Experimental results demonstrate that implementing the proposed technique with the adapted pre-trained models surpasses existing state-of-the-art methods.This conclusion is drawn from a comprehensive evaluation involving metrics, charts, and graphs.Notably, the results reveal that employing the technique with the MobileNetV2 pre-trained model achieves the highest detection accuracy rate, approximately 95%, while requiring fewer training parameters, leading to expedited training times.