Personalization of Candide 3D Model for Human Computer Interfacing

Putria Febriana and Władysław Skarbek
Warsaw University of Technology, Dept. of Electronics and Information Technology,
Nowowiejska 15/19, 00-665, Warsaw, Poland

This paper refers to the personalization of human head 3D model to be used as an element of human-computer interface (HCI). The distinctive personal features are extracted from each camera frame using Candide 3D model. Namely, for the selected face parts, 14 shape deformation units are specified. By nonlinear least square method, the parameters for the shape deformations are identified.The orthographic projection of Candide 3D points corresponding to on-line detected 68 facial landmarks is used to calculate error function. Statistical analysis proves that the shape deformation coefficients in majority of cases exhibit low variance and therefore they can be applied to adjust the Candide model in order to get the personalized model. The personalized model is crucial for improvements of other HCI application like face expression recognition.

Keywords: Face 3D model, shape deformation units, model personalization, face detection, facial landmarks detection, nonlinear least square optimization.