Deepfake Image Forgery Detection using Local Feature Descriptors
S Dienash, T. Swapna · 2024
Deepfake Image Forgery,generates fake images after manipulating the original source images. This increases the risk of losing trust in images or videos. This paper proposes an efficient methodology to detect Deepfake images from real images. The methodology involves extracting Local Feature De scriptors(Local Binary Pattern (LBP),Frequency Decoded Local Binary Pattern (FDLBP),Binary Gabor Pattern (BGP)). The fusion of these features are fed to the DenseNet to classify the image as fake or real. The model uses Convolutional Neural Network (CNN) architecture that utilizes all three Local Feature Descriptors that extract features of the images. The model has an accuracy of $\mathbf{9 3. 6 5 \%}$ which is performing better than Deep learning models.