Unlearnable example with face images

• Main Author: Peng, Haohang
• Other Authors: Alex Chichung Kot
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2025
• Subjects: Computer and Information Science; Unlearnable example; Deep learning
• Online Access: https://hdl.handle.net/10356/182488
• Institution: Nanyang Technological University

This thesis focuses on the critical issue of image protection, with a particu- lar emphasis on safeguarding face images by introducing perturbations to these images. Deep learning models have demonstrated remarkable potential to drive significant advancements across various fields, including the generation of new images and the enhancement of object detection capabilities. However, alongside these advancements, there are inherent risks to personal privacy that cannot be overlooked. These risks arise through multiple avenues, especially in light of the rapid development of generative AI technologies such as stable diffusion, which empower individuals to create images using just a few reference pictures. To address this pressing problem, this research delves into the intricate rela- tionship between unlearnable examples (UEs) and deep learning models. We conduct a thorough analysis of how UEs can be effectively applied within the realm of generative AI. Furthermore, we extend our investigation to the use of UEs in object detection, aiming to ensure that models are unable to accurately detect or interpret these images, thereby enhancing privacy protection measures in the process.