Year: 2026 | Month: May | Volume: 13 | Issue: 5 | Pages: 631-635
DOI: https://doi.org/10.52403/ijrr.20260559
Deep Fake and Image Forgery Detection using Machine Learning
Pamula Kamakshi1, Veeragani Harika1, Kowsika Paladugu2, Pentapati Karthikeya3, Brahma Teja Rayapaneni4, Sai Manaswy Manukonda5
1,2,3,4,5Department of Information Technology, Dhanekula Institute of Engineering and Technology College, JNTUK, Vijayawada, India.
Corresponding Author: Veeragani Harika
ABSTRACT
The increasing prevalence of advanced digital image editing capabilities makes it harder to detect image fakes using traditional methods and creates major problems for digital forensics, journalism and courtroom proceedings. In this article we explain Image Guard AI: A multi-module hybrid deep learning framework that combines a highly optimised MobileNet Convolutional Neural Network (CNN) with multiple traditional machine learning classifiers (such as Support Vector Machine, Logistic Regression, Decision Tree, K-Nearest Neighbours and Random Forest) using an ensemble model to support each other. The proposed use of Error Level Analysis (ELA) pre-processes images to help increase the noise created by compressing the images, which is a by-product of local image edits (or fakes) made to them. MobileNet generates 128-dimensional representations of the ELA image that are provided to the individual machine learning classifiers. The output or predictions from all classifiers are combined through a soft voting ensemble into a final weighted classification prediction using confidence scores from the classifiers. The experimental evaluation of the proposed ensemble voting classifier using the benchmark forgery detection database demonstrated that Image Guard AI achieved 97.2% accuracy, 97.5% precision, 96.9% recall, and an AUC of 0.993, which was consistently higher than any of the individual classifiers. The implementation of this system has produced a web application (Flask-based) with a real-time user interface for detecting fake images. The results found with the proposed methodology validate that the combination of deep feature extraction using mobile networks with ensemble builds provide for effective and reliable detection of digital image forgeries/general fakes and are general across all images regardless of content or makeup.
Keywords: image forgery detection, deep learning, MobileNet, Error Level Analysis, ensemble learning, SVM, hybrid classifier, digital forensics
[PDF Full Text]