Enhancing Digital Crime Investigation Through Deep Learning-Powered Deepfake Detection
Navaneeth Bhaskar, Priyanka Tupe-Waghmare, Shravya Ganesh, Dhanya Vinay, K Leena, Dayana Dsouza · 2024
The increase of deepfake videos compromises the integrity of information delivery since they can be used to spread misinformation, manipulate public opinion, and cause conflict. Detecting deepfake media is a difficult task, especially in forensic circumstances where accuracy is very important. In this study, we developed an advanced deep-learning model specifically for the automated detection of deepfake images and videos. Our approach uses convolutional neural networks (CNNs) to distinguish between real and fake media. In addition, to improve the prediction accuracy, we employed hybrid deep learning methodologies, which include both CNN and Support Vector Machine (SVM) classifiers. Our proposed CNN-SVM hybrid model achieved an accuracy of more than 95%. Furthermore, the algorithm was trained to predict a person’s age and gender from the image, which is significant in digital crime investigation. This work is an essential contribution to ongoing attempts to fight against deepfake content and increase trust in multimedia content. Implementing such models will discourage people from creating and spreading deepfake images and videos. As a result, people’s privacy is secured, digital media integrity is maintained, and the credibility of digital media ecosystems is preserved.