Deep learning for facial expression editing

• Main Author: Wu, Rongliang
• Other Authors: Lu Shijian
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/168477
• Institution: Nanyang Technological University

In this day and age of digital media, facial expression editing, which aims to transform the facial expression of a source facial image to a desired one without changing the face identity, has attracted increasing interest from both academia and industrial communities due to its wide applications in many tasks. Automatic facial expression editing has been explored extensively with the prevalence of generative adversarial networks in recent years. Although some research works have been reported and achieved very promising progress, the task of facial expression editing is still facing four major challenges, including the unsatisfactory editing quality issue, the constrained data annotation issue, the limited controllability issue and the multi-modality issue. This thesis focuses on the above-mentioned challenges in facial expression editing task and introduces several novel deep-learning-based techniques to alleviate the corresponding challenges. Extensive experiments show that the proposed approaches achieve superior performance in facial expression editing.