Facial expressions using a Quadratic Deformation Model: Analysis and synthesis
C. Tay, Mohammad Obaid, Ramakrishnan Mukundan, Andrew Bainbridge-Smith · 2009
In this paper we proposes a novel method to generate facial expressions on two-dimensional sketch-models and images. The six facial expressions (smile, sad, angry, disgust, fear, and surprise) are represented using a set of quadratic deformation parameters defined on muscle-based regions. Quadratic deformation parameters are obtained in a preprocessing stage when motion capture data for several points on a human face is used. A general algorithm for mapping the transformation parameters onto two-dimensional images to generate the corresponding expressions is developed. We give the implementation aspects related to expression generation, and describe various stages involved in the process. Two user case studies, image-based and video-based, have been taken to study the performance of the developed method. Among 30 participants, 52.96% can correctly identify the image-based static expressions and 58.67% for video-based animated expressions. The correct response for facial expressions improves when the process of transformation is showing to the audience. Several limitations of the project are indentifled and discussed; however, the overall result is still acceptable. Further development directions have been suggested for future research.