Department: Computer Science & Engineering
Research Institute Affiliation: Agile Center for Visual Computing
Faculty Advisor(s): Ravi Ramamoorthi

Primary Student
Name: Muhammad Ahmed Riaz
Email: mriaz@ucsd.edu
Phone: 858-281-1761
Grad Year: 2016

Reconstructing 3D human face from a single image is inherently an ill-posed problem, which requires additional cues to obtain a solution. For instance, such a solution can exploit the fact that human faces are constrained to a small subspace of all possible shapes and have limited variation. In existing literature, two main classes of solutions exist for this problem; parametric face models or shape from shading. We propose to combine the two said classes of solutions to get a more accurate depth recovery. A parametric face model is used to capture the global details of the face, which is followed by shape form shading phase which adds high frequency local details to the depth. Existing methods also ignore the cast shadows on face due to non-frontal lighting. We lift this assumption by estimating these shadows in our optimization using a Pre-computed Radiance Transfer system. Inclusion of cast shadows improves the results under novel lighting conditions. This system provides a low cost face modeling solution. The requirement of just a single image means a normal mobile phone with no additional sensors can be used to generate 3D face models.

Industry Application Area(s)
Software, Analytics | Computer Vision

Related Links:

  1. http://cseweb.ucsd.edu/~mriaz/images/3D_face_phong2.mp4

« Back to Posters or Search Results

Contact:   researchexpo@soe.ucsd.edu   (858) 534-6068