SPECIFIC FACE 3D RECONSTRUCTION

BASED ON SNAKE MODEL

WANG Xuanyin  PAN Feng  XIANG Guishan  LIANG Dongtai

(State Key Laboratory of Fluid Power Transmission and Control, Zhejiang University, Hangzhou 310027)

Abstract: A reconstruction system for special human face is designed which is based on the binocular stereo vision. A 3D reconstruction method for special human face based on Snake model is presented. The stereo image pair is gathered with a calibrated camera, and the face area is segmented by skin detection method. After the image pair is processed with distortion correction and epipolar line alignment, and the face area is selected and normalized, the correlation matching algorithm adopts pyramid structure which can balance well between error matching points and precision. An energy minimization function for human face is proposed which takes into account the order matching constraint, the continuity constraint, the curvature estimation of every point on curve and the correlation constraint condition. The function improves greatly the accuracy of disparity extraction, resolves the stereo matching difficulty in stereo vision area, restores successfully the 3D data of human face and realizes the reconstruction of special face. Experiments show that the algorithm which adopts the disparity extraction based on Snake model and fairing handling and texture map-ping is realistic in the reconstruction of main face part, except the deformation along partially face contour. The algorithm is effective and practical, which can restore the 3D data of special face without structured light and manual intervention.

Key words: Face 3D reconstruction Stereo match Pyramid structure Correlation calculation Snake model

CLC No: TP391

新世纪优秀人才计划资助项目(NCET-04-0545). Received 20060704, received in revised form 20070604