Ensemble-Based Deep Learning Framework for Detecting GAN-Generated Fake Images
Harsha Velpumadugu, K. Swaraja, Prathap Velthapu, Srilakshmi Aouthu, Ajay Gokhle, P A Harsha Vardhini · 2025
Deepfake images, made through modern AI technologies like Generative Adversarial Networks(GANs), making it challenging to differentiate between real and manipulated content. These type of deepfake images can lead to various significant problems like circulating false information or harming individual's privacy. Due to the advancement in new technologies, decent methods for determining real images from fake ones are necessary. The proposed work aims on building a system called Ensemble-based deep learning framework for detecting GAN-Generated fake images. In this ensemble based approach, it combines features from three different pre-trained models namely DenseNet121, VGG16, and EfficientNetB4 along with a handcrafted Grey Level Co-occurrence Matrix(GLCM) to enhance accuracy. These models each process the image individually and combines their outcome via global average pooling and concatenation for final image classification. This proposal come up with a powerful tool to resist the misuse of deep fake technologies. The designed model is tested on multiple datasets and demonstrates encouraging results in identifying deepfake images and it outperformed 98 % accuracy in the automatic detection of GAN generated synthetic images.