Technical Solutions

This chapter familiarizes the reader with deepfake technologies. Thus, I will provide a synthesis of the most recent research on deepfake technology, its legal implications, the use of Generative Adversarial Network models for creating deepfakes, AI methods for identifying and combating them, and the vulnerabilities of facial recognition systems in the context of deepfakes. In this chapter, I proposed a new algorithm for detecting images with deepfake content. This algorithm aims to avoid using AI methods, which have the disadvantage of requiring large data volumes for training. The proposed algorithm uses a correlation between Euclidean distance and head position. These indicators are compared to those from a reference image, known with certainty to be authentic. I have used Google Vision API in the DeepFake Algorithm (DFA) to identify key facial points for these indicators. Subsequently, I conducted a total of 40 tests: 20 on original selfies and 20 on deepfake images generated using my selfies. My algorithm correctly detects deepfake images with a recall percentage of 100%, having an accuracy of 75%. At the same time, the algorithm does not detect all the authentic images. The calculated F1-score is 80%. For this reason, the DFA requires further improvements but represents a starting point for research in the field of deepfakes. The algorithm makes a theoretical contribution by proposing a method for detecting deepfakes based on the geometric analysis of facial keypoints. The proposed algorithm brings practical implications in fields such as cybersecurity, biometric authentication, unauthorized use of private data, and the fight against misinformation.