Estudio metodológico de las herramientas tecnológicas actuales para detectar y reconocer DeepFakes, abordando la creciente amenaza de la manipulación de contenido multimedia generado por inteligencia artificial.

Abstract

This thesis addresses the growing threat of DeepFakes in the digital age, where manipulating multimedia content generated by artificial intelligence can make distinguishing between accurate and false information difficult. The overall objective of this study was to identify, analyze, and evaluate the technological tools used to detect and recognize DeepFakes to address this challenge effectively. A workflow has been designed to clarify the processes and steps to be followed, thus allowing the development of a solid and well-founded methodology. The choice of the AUC metric as the evaluation criterion has made it possible to globally measure the discriminative capacity of the tools globally, ensuring an objective and fair comparison. The results obtained from comparing and evaluating five DeepFakes detection tools have revealed that Capsule stood out with the highest scores in different studies, excelling especially in the FaceForensics++ dataset. Furthermore, areas of improvement and challenges in DeepFakes detection have been identified, highlighting the importance of considering different scenarios of fake video generation and the adaptability of the tools to future manipulation techniques. This research provides a valuable contribution to the field of DeepFake detection, offering a solid foundation for future research in this area. The results and conclusions obtained will significantly interest researchers, experts, and individuals interested in DeepFake detection and mitigation. Implementing these detection tools and their continuous improvement will allow preserving the integrity and veracity of digital content in the context of rapid technological evolution. In this way, it is expected to address the threat of DeepFakes and promote trust in information today.