What is it about?
This work proposes and studies a method to enhance the robustness of facial recognition systems. An evolutionary approach is used to explore the latent space of GANs to generate synthetic images of human faces that deceive a facial recognition model, using two faces of different real people as input. Results allow identifying the weakness of the facial recognition systems, which recognize the generated face as matching both input identities.
Featured Image
Photo by Malte Helmhold on Unsplash
Why is it important?
Facial recognition systems are widely used in a variety of day to day real applications. In general they are trained with faces of real people. Understanding their weaknesses, in particular when fed with artificially generated faces, is an important step towards more robust recognition approaches.
Perspectives
The research contributes to detecting biases existing in both generation and recognition models, as well as to exposing their weaknesses when used in real life applications.
Benjamín Machín
Universidad de la Republica Uruguay
Read the Original
This page is a summary of: Multi-target evolutionary latent space search of a generative adversarial network for human face generation, July 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3520304.3533992.
You can read the full text:
Contributors
The following have contributed to this page
