International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

A Facial Deepfake Detection Approach using CNN-based Models, Swin Transformer and Classifier Fusion

Authors :

Alireza Honardoost¹ Mahdie Rahmati² Babak Nasersharif³

1- Faculty of Computer Engineering K. N. Toosi University of Technology 2- Faculty of Computer Engineering K. N. Toosi University of Technology 3- Faculty of Computer Engineering K. N. Toosi University of Technology

Keywords :

Deepfake Detection،Swin Transformer،CNN،VGG،ResNet،Classifier Fusion

Abstract :

Facial deepfake detection focuses on identifying manipulated images and videos generated by deep learning methods such as Generative Adversarial Networks (GANs) and autoencoders. Detection methods range from analyzing facial and physiological features to using deep learning models like Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). In this paper, we propose a method based on combining CNNs and ViTs in the level of their decisions for classifying real or fake images. This combination is performed by different static classifier fusion methods on the probability outputs of classifiers. The used CNN models are VGG16 and ResNet50, and the used ViT model is Swin-Tiny. To evaluate the proposed method, the VidTIMIT and DeepfakeTIMIT datasets are employed. The results show that combining fine-tuned models leads to improved accuracy across dataset, outperforming the individual models.