Exploration of Generative Neural Networks for Police Facial Sketches
Nerea Sádaba-Campo, Hilario Gómez-Moreno · Big Data and Cognitive Computing · 2025
This article addresses the impact of generative artificial intelligence on the creation of composite sketches for police investigations. The automation of this task, traditionally performed through artistic methods or image composition, has become a challenge that can be tackled with generative neural networks. In this context, technologies such as Generative Adversarial Networks, Variational Autoencoders, and Diffusion Models are analyzed. The study also focuses on the use of advanced tools like DALL-E, Midjourney, and primarily Stable Diffusion, which enable the generation of highly detailed and realistic facial images from textual descriptions or sketches and allow for rapid and precise morphofacial modifications. Additionally, the study explores the capacity of these tools to interpret user-provided facial feature descriptions and adjust the generated results accordingly. The article concludes that these technologies have the potential to automate the composite sketch creation process. Therefore, their integration could not only expedite this process but also enhance its accuracy and utility in the identification of suspects or missing persons, representing a groundbreaking advancement in the field of criminal investigation.