Controllable face de-identification In constrained environments

• Main Author: Lim, Guo Quan
• Other Authors: Liu Yang
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/166024
• Institution: Nanyang Technological University

This project proposes a workflow and method to perform face de-identification while allowing the ability to edit a single binary facial attribute. We are also able to maintain another single binary facial attribute. In order to do so, two previously proposed frameworks will be combined into one single workflow, capitalizing on the strengths of each individual framework. They will complement each counterpart’s strengths. We will first test each individual framework and its efficacy when it comes to the task of face de-identification. The reason for this is because each framework was invented without the intention of face de-identification. We want to further prove that our proposed workflow has increased efficacy of face de-identification compared to each individual framework. Since these two frameworks can be arbitrarily ordered in our workflow, we will test each single workflow to find which one works the best in achieving our project goals which we will further state in this report. This project report is structured such that we will introduce the project title at hand, and break down the title further into two components and understand what these components mean. This provides us with context for the rest of the report and our motivations for this project. We will then expound our project goals, and state the end goal in which we hope to get and achieve. Our final edited workflow image results will be placed in Appendix 2 for further viewing and qualitative assessment.