A photometric sampling method for facial shape recovery

The authors propose a photometric method to recover facial shape that is consistent with expected facial proportions. The method borrows ideas from photometric sampling, a technique that estimates shape from continuous variations of a light source around a single circular path. This approach aims at...

Full description

Saved in:
Bibliographic Details
Published in:Machine vision and applications 2016-05, Vol.27 (4), p.483-497
Main Authors: Hernandez-Rodriguez, Felipe, Castelan, Mario
Format: Article
Language:English
Subjects:
Communications Engineering
Computer Science
Estimating
Facial
Image Processing and Computer Vision
Light sources
Lighting
Luminance
Mathematical analysis
Matrix algebra
Matrix methods
Networks
Original Paper
Pattern Recognition
Photometry
Sampling
Trigonometric functions
Vision systems
Zenith
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The authors propose a photometric method to recover facial shape that is consistent with expected facial proportions. The method borrows ideas from photometric sampling, a technique that estimates shape from continuous variations of a light source around a single circular path. This approach aims at enriching photometric information by including variations of the light source along its zenith angle. To this end, a luminance matrix describing lighting response along both azimuth and zenith angles of the light source is built for each pixel. A method based on fitting sine functions onto the singular vectors of the collected luminance matrices is proposed for estimating a surface normal map. The estimated surface normals are later refined to maximize a facial proportion criterion and finally be integrated. Experiments demonstrate that our approach successfully approximates 3D face shape while preserving facial proportions within the limits of expected depth.
ISSN:0932-8092
1432-1769
DOI:10.1007/s00138-016-0755-9