Abstract
Compared with other methods, such as using a 3D laser scanner, multi-view images etc, a single view-based 3D face synthesis method is the most appropriate to make supervision free, resulting in a fast performance aftermath. However, the existing single view-based face synthesis algorithms have problems in depth estimation, systemic automation, interoperability, and being able to deal with complex backgrounds. Aimed at resolving the existing problems, a single view-based face synthesis algorithm has been developed, which enables 3D face synthesis from an arbitrary head-and-shoulder image with a complex background. Compared with other existing single view-based algorithms, the developed one has many advantages. Unlike some [Feng00a] [Val0l] that require user interaction, the developed algorithm can implement automated face synthesis; compared with those single view-based algorithms only capable of coping with faces in the neutral state [Hu04] [Kuo02], the developed algorithm is undoubtedly suitable for wider range of head pose; in contrast to those single view-based methods [Fen00a] [Ho01] [Kuo02] [Hu04] targeting only pure face images, our system can tackle those face images with more complicated background. Moreover, a number of technical improvements for face detection, chin extraction and face model adaptation etc, have also been made in the developed system. In light of the configuration of the developed single view-based face synthesis system, the thesis falls into five chapters, apart from the Introduction and Conclusion chapters. Chapter 2 reviews the current state of the art technologies in 3D face synthesis and their applications, leading to the motivation of the thesis. Chapter 3 and 4 are dedicated to spatial face detection and facial feature extraction, respectively, followed by Chapter 5 that focuses on face model adaptation and 3D face texturing. Since many technologies, such as 3D face rendering and 3D model-based video coding etc, demand a seamless coupling of face synthesis and facial animation techniques. Chapter 6 conducts a technical introduction to facial animation from its principle to implementation. Finally, the development of this single view-based 3D face synthesis system can benefit many human machine applications, such as pose invariant face recognition and 3D model-based video coding.