A Component-based Face Synthesizing Method

The active appearance models (AAM) is a popular tool in object tracking. An AAM is featured by its integrated modeling of deformations in both shapes and textures. Therefore, in addition to the object tracking, AAM is also a good visual synthesizer. The other strength of the AAM is its compact representations for the geometries and textures of synthesized objects. By training with the principal component analysis method, the AAM parameterizes the shape and the texture of each synthesized object simply with a linear combination of eigen-shapes and eigen-textures, respectively. This paper presents a novel video-driven face synthesizing method which tracks the faces of a person on video frames and synthesize novel faces using geometries of individual facial components, such as eyes, noses, and mouths, of other persons. To this end, we propose the component-based active shape models (ASM) for synthesizing each facial component. One major prominent feature of the proposed method is that a rich number of novel facial expressions can be synthesized by combining different facial components from different persons on the synthesized faces. No further retraining process for the AAMs or ASMs is required for synthesizing these novel facial expressions. The experimental results show that the proposed method can accomplish interesting and vivid facial synthesis and exhibits its high potential in many practical applications.