Análisis de Técnicas de Detección de Imágenes Sintéticas

Abstract

This research addresses the study and analysis of the current state of detectors for synthetic images generated by diffusion models. Through a methodology that includes the study of previous generative models and the application of advanced detection techniques such as binary classifiers and frequency spectrum analysis, this work explores the efficacy of current methods and their applicability to diffusion models. The results reveal that, although the classifiers are effective within the datasets with which they were trained, they face significant limitations in generalizing to new models, highlighting the need for more robust and generalizable tools. The conclusions emphasize the urgency of continuing research in this field, given the serious implications of deepfakes and unregulated synthetic content. This study not only validates the possibility of adapting existing techniques to new generative models but also provides a starting point for the development of practical solutions in the fight against cybercrime, specifically in the detection and prevention of illegal and harmful content. The tools developed from this research could be essential for security forces and other entities dedicated to combating the pernicious effects of deepfake technology.