Abstract
Digital forensics is crucial in investigating and analyzing digital evidence, including images, to uncover potential crimes and identify manipulated or fraudulent content. A significant challenge in digital forensics is distinguishing between authentic and manipulated images. This paper applies a popular pre-trained deep learning model, VGG16, for feature extraction in real vs. fake image detection and various machine learning (ML) models as classifiers. The feature extraction part until the flatten layer of the VGG16 model is utilized to extract discriminative features that capture high-level image representations comprising local and global patterns. ML models, including Random Forest, Logistic Regression, Decision Tree, and k-Nearest Neighbors, are explored to distinguish between real and fake images. ML models trained on a labeled dataset, encompassing a wide range of authentic and manipulated images, to learn the underlying patterns and correlations. The experimental results demonstrate that a combination of VGG16-based features and the powerful learning capabilities of the RF provides the highest detection accuracy of 78.56 %, thereby enhancing the efficacy of forensics investigations in detecting image forgeries and ensuring the integrity of digital evidence.
Recommended Citation
Alashjaee, Abdullah Mujawib
(2025)
"Machine Learning Approach for Fake Image Forensics,"
University of Bisha Journal for Basic and Applied Sciences: Vol. 1:
Iss.
1, Article 5.
Available at:
https://ubjbas.ub.edu.sa/home/vol1/iss1/5