DL-FID: Transfer Learning-Based Fake Synthetic Image Identification

The rapid progress of deep learning technologies has led to generation of realistic fake images, expressing concerns about their improper use in domains like misinformation, identity theft, and digital forgery. This paper explores a transfer learning-based approach for identifying fake synthetic images, exploits pre-trained deep neural networks to enhance detection accuracy and efficiency. This study uses transfer learning to fine-tune models on large dataset such as CIFAKE, and distinguishes between the selected datasets of real and fake images. We have proposed DL-FID: A deep learning-based model with EfficientNetB0 as baseline model. The EfficientNetB0 Model, known for its superior feature extraction capabilities, was chosen. This study conducted a detailed analysis to understand the impact of various components on the model’s performance, and revealed the model’s ability to capture differences between real and fake images.