Output details

Motion-Sensitive Anchor Identification of Least-Squares Meshes from Examples

Significance: Facial Animation plays a critical part in conveying emotion and increasing the realism of virtual characters. However, facial rigging is challenging, requiring the use of a large number of pose-able feature points which allows an animator to define a unique facial pose. For example in the film ‘The Incredibles’, as many as 1700 feature points were needed for each face. This practice is tedious, time-consuming and therefore costly, given that most facial poses are frequently repeated, particularly during speech.

Originality: In this paper we have developed a method to reduce the number of these feature points using a novel robust statistical approach: Clustered Teleconnection Analysis. This method demonstrably improves on existing approaches for key point simplification by giving a better “bang for buck” for each additional feature point added. It is also useful in other application areas of computer graphics, such as selecting representative light samples in global illumination solutions or intelligent selection of key poses for the definition of deformation spaces.

Rigour: We demonstrate our approach significantly improves on previous methods for key point selection. We apply this technique to other models where we have a set of example poses, demonstrating a significant reduction in error over these previous methods. We provide performance results which can be used to guide parameter selection.