Recovering 3-D Face Models using a Hybrid Structured – Light Stereo Matcher

A. Naftel and Z. Mao (UK)

Keywords

stereo, 3D, structured-light, shape, face models

Abstract

This paper describes a hybrid structured-light assisted stereo (S-LAS) matching algorithm for acquiring 3D facial models. We show that capture of an additional structured-light stereo pair is useful for partitioning stereo images into corresponding sub regions. Regular image partitions can be exploited to correct for scale distortion and restrict the potential correlation search space when determining point correspondences. In experiments using calibrated images of target points with known ground truth, we have obtained average RMS matching errors of 0.2 pels and 3D reconstruction errors of 0.3mm. In a study using human face images, it is shown that the S-LAS algorithm is capable of increasing the density and quality of point correspondences compared with a conventional two-frame stereo approach.

Important Links:



Go Back


IASTED
Rotating Call For Paper Image