Deepfake detection and classification using local surface geometrical features

The rise of deepfake technology presents a critical challenge to the integrity of digital media, prompting the need for advanced forgery detection techniques. This paper proposes a novel approach for detecting deepfake image forgery by integrating local surface geometry analysis into the detection process. Leveraging the inherent alterations in local surface geometry caused by deepfake generation, our method extracts and merges this information with the original image, enhancing the discriminative features essential for accurate identification of manipulated content. Using classification networks such as Resnet, Densenet, and Efficient, our method achieves a remarkable accuracy rate of above $98 \%$ in correctly identifying deepfake images for the classification models. This high accuracy underscores the effectiveness of our approach in detecting sophisticated deepfake content. By combining local surface geometry analysis with deep learning-based classification, our method represents a significant advancement in the field of deepfake detection, offering a promising solution to combat the proliferation of manipulated media and preserve the authenticity of digital imagery.