Detecção de imagens geradas por inteligência artificial: um estudo sobre técnicas e desafios

Abstract

This research presents a Systematic Literature Review (SLR) on methods for detecting images generated by artificial intelligence, focusing on approaches published between 2022 and 2025. A total of 44 articles were analyzed following the PRISMA protocol, allowing the mapping of the state of the art, the identification of technical advances, and the recognition of persistent challenges. The findings reveal a significant growth in scientific production in the field, with emphasis on architectures based on Convolutional Neural Networks (CNNs), Transformers, and hybrid models, as well as techniques related to cross-domain generalization, latent space exploration, and multi-scale attention mechanisms. In addition to technical aspects, the SLR also identified ethical and social concerns, including misinformation, privacy risks, and the need for greater transparency and regulatory frameworks. In summary, the review confirms that synthetic image detection is a rapidly evolving research area, still marked by challenges but showing significant advances that highlight its scientific and social relevance.