45. BEYOND THE FACE BOX: INCORPORATING HEAD SHAPE INTO FACE IDENTIFICATION

Department: Computer Science & Engineering
Faculty Advisor(s): Serge Belongie | David Kriegman

Primary Student
Name: Eric M Christiansen
Email: echristi@ucsd.edu
Phone: 760-814-1394
Grad Year: 2013

Student Collaborators
Iljung Sam Kwak, iljung.kwak@gmail.com | Andrew Ziegler, andrewzieg@gmail.com

Abstract
Current methodologies for face identification operate on all pixels within a face box, and do not explicitly recognize which pixels in the box belong to the face and which are background. In this paper, we study the additional discriminative power provided by binary face / background masks. We perform the study on the MultiPIE dataset of condition-varying face photographs, for which we have computed high quality masks. For the analysis, we introduce two shape features, one based on the BRIEF descriptor and another on curve parametrizations, and find face shape alone is discriminative in non-frontal poses, and augments standard full-box techniques.

« Back to Posters or Search Results