Building 3-D Human Face Models from Two Photographs

  • Authors:

  • Qian Chen;Gérard Medioni

  • Affiliations:

  • Integrated Media Systems Center, University of Southern California, Los Angeles, CA 90089, USA;Integrated Media Systems Center, University of Southern California, Los Angeles, CA 90089, USA

  • Venue:
  • Journal of VLSI Signal Processing Systems - Special issue on multimedia signal processing
  • Year:
  • 2001

Quantified Score

Hi-index 0.01

Visualization

Abstract

This paper presents a working system for building 3-D human face models from two photographs. Rather than using expensive 3-D scanners, we show that frontal face models can be faithfully reconstructed from two photographs taken by consumer digital cameras in a totally non-invasive setup. We first rectify the image pair so that corresponding epipolar lines become coincident, by computing a dual point transformation. We then address the correspondence problem by converting it into a maximal surface extraction problem, which is then solved efficiently. The method effectively removes local extrema. Finally, a Euclidean reconstruction is achieved with the help of a novel factorization method for perspective cameras. Most of the computational steps are conducted in projective space. Euclidean information is introduced only at the last stage. This sets apart our system from the traditional ones which begin with metric information by using carefully calibrated cameras. We have collected a bank of face pairs to test our system, and are satisfied with its performance. Results from this image database are demonstrated.