Enhanced Detection of GAN‐Generated Image Forgeries Using Error Level Analysis and Deep Learning Models

ABSTRACT In the digital age, distinguishing fake images has become increasingly difficult, particularly with the rise of generative adversarial networks (GANs) that produce highly realistic images, especially faces, which are often indistinguishable from real photographs. Despite numerous efforts by researchers to develop effective GAN detection systems, challenges persist due to the lack of comprehensive datasets and effective models. To address these challenges, we propose a forgery image detection method that combines error level analysis (ELA) with convolutional neural networks (CNNs) for detecting GAN‐generated images. Our approach integrates the ELA technique with transfer learning, leveraging its better architectural design to extract hierarchical features for efficient classification. The ELA technique identifies inconsistencies in images, which are then processed using multiple transfer learning models, including MobileNetV2, DenseNet121, VGG16, VGG19, Xception, InceptionV3, ResNet‐50, EfficientNetB0, Custom, and Ensemble models. Among these, EfficientNetB0 and DenseNet121 demonstrated high accuracy, with EfficientNetB0 outperforming DenseNet121 in detecting forgeries due to its more efficient feature extraction and classification capabilities. Another key contribution of our work is the introduction of the novel HFD‐8000 dataset, which addresses the scarcity of GAN‐specific datasets in the field. This dataset contains 8000 images, both original and GAN‐forged, sourced from various origins with proper consent, and is publicly available on Mendeley Data. By providing comprehensive data, the HFD‐8000 dataset fills a critical gap in GAN forgery research and offers a solid foundation for future studies in the domain. Our proposed model, incorporating ELA with advanced deep learning techniques, achieves 96.37% accuracy and high performance in forgery detection, providing an effective solution to the growing challenge of image manipulation in the digital age.