Year: 2025
Type: article
Abstract: Deepfake technology threatens the credibility of digital media increasingly, allowing for the creation of highly realistic but false videos and images. Proper detection of deepfake content is required to help preserve trust and avoid exploitation. In this paper, a new deep learning method that uses ResNet18 with pre trained weights is introduced to achieve higher feature extraction and classification accuracy in detecting deepfakes. A sub-sample of real and forged images undergoes resizing, normalization, and data augmentation algorithms to provide model generalization. Model optimization is obtained using a fully connected layer, dropout regularization, and sigmoid activation function to enable binary classification. Training is done using the Binary Cross-Entropy loss function and optimized using the Adam optimizer to obtain quicker convergence. Model accuracy and ROC-AUC on test and validation sets are used as metrics to measure performance. The proposed framework illustrates a scalable, efficient, and effective means to identify deepfakes with real-world implications in social media monitoring, digital forensics, and cyber security.
Authors: Kalpana Kumari, Prof. Nitesh Gupta
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4414625996
FWCI:
Cites:
Cited by:
Related to: