Image Forgery Detection Using Convolutional Neural Networks
Ayesh Meepaganithage, Suman Rath, Mircea Nicolescu, Monica Nicolescu, Shamik Sengupta
Abstract
With the advancement of technology, the ability to forge realistic images has become increasingly accessible, and this has led to a significant challenge in the digital forensics field. A lot of research has been conducted in this area to address this challenge. Traditional image forgery detection methods are time-consuming and slow. Therefore, deep learning methods have played a vital role in image forgery detection. In this research, we investigate the use of convolutional neural network models for image forgery detection and evaluate their performance. From the experimental results, it can be seen that ResNet models have demonstrated high performance in detecting forged images accurately. Out of all the convolutional models we evaluated, the ResNet-101 model obtained the best results, with 93.46% accuracy, 95.08% precision, 92.10% recall, 94.91% specificity, and 93.57% F1-score. ResNet-101 model correctly identified 621 out of 650 authentic images and 584 out of 650 forged images.