Advanced deepfake detection using hybrid CNN-MLP models with feature extraction from facial landmarks

Abstract

The study presents a novel hybrid deepfake detection model that combines Convolutional Neural Networks (CNN) with Multi-Layer Perceptron (MLP) to enhance the accuracy of deepfake detection. Deepfakes are fabricated multimedia materials created using deep learning algorithms, posing a significant threat to the integrity of online information. This research utilizes facial landmarks as a feature extraction method to enhance the hybrid CNN-MLP model's ability to detect geometric inconsistencies and facial distortions. The proposed model is evaluated against traditional CNN and Long Short-Term Memory (LSTM) networks. Results show that the hybrid CNN-MLP model achieves superior performance, with an accuracy of 98.99%, precision of 98.99%, and F1-score of 98.99%, outperforming both CNN and LSTM. This hybrid approach demonstrates robust detection capabilities across various deepfake generation techniques, offering an effective solution for mitigating the spread of falsified digital content.