Enhanced Image Tampering Detection using Error Level Analysis and CNN
Ramesh Gorle, Anitha Guttavelli
Abstract
This paper introduces a novel approach to image tampering detection by integrating Error Level Analysis (ELA) with a Convolutional Neural Network (CNN). Traditional forensic methods, such as ELA and Residual Pixel Analysis (RPA), often struggle to detect subtle or advanced manipulations in digital images. To address these limitations, this method leverages ELA to highlight compression-induced variations and CNN to extract and classify spatial features indicative of tampering. The dataset, consisting of both authentic and tampered images, was preprocessed to generate ELA representations, which were then used to train a CNN model designed to distinguish between authentic and manipulated regions. Extensive experimentation was performed on the CASIA v2.0 dataset, demonstrating significant improvements in detection accuracy, precision, and recall. The proposed framework achieved a detection accuracy of 96.21%, outperforming established deep learning models such as VGG16, VGG19, and ResNet101. These results underscore the potential of combining ELA and CNN in advancing image forensics, offering a robust solution to ensure the integrity of digital content in an era of sophisticated digital manipulation.