Facial Expression Synthesization Using Geometrical Warping

Facial expression transfer has long been an important topic in computer graphics and vision, driven by applications in character animation, computer games, advertising and more recently, healthcare. However, how to make the synthesized facial expression realistic remains a challenge due to the complexity of human facial anatomy and our inherent sensitivity to facial expression. The current thesis describes a system that automates facial expression transfer. The process consists of two main parts: feature points detection and geometrical deformation. Feature points detection is achieved by using Active Shape Model (ASM). Inaccurate feature locations in ASM are subsequently corrected by recursively searching for the maximum gradient in local area. This is followed by geometrical deformation in which geometrical warping parameters estimated between feature points from the source faces are applied to the target face. The experimental results show that this method works well on still frontal facial expressions.