Evaluation of Landmark-Based Face Recognition Using Bidimensional and Tridimensional Regression

• Activity Number: 253
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Kendra Schmid*+
• Companies: University of Nebraska Medical Center
• Keywords: bidimensional regression ; tridimensional regression ; face recognition ; shape analysis ; landmark data

Abstract

Face recognition is a nonintrusive method of identifying humans based on their unique facial features. Bidimensional and tridimensional regression are two statistical approaches that can be used to measure similarity between two objects represented by landmark coordinates. Both types of regression can be used as a tool for mapping points and computing the degree of similarity between two configurations of points, given in a set of matching coordinates. Three transformations are used to investigate which has the best ability to match faces using two-dimensional and three-dimensional landmark coordinates. Multiple images of thirty two subjects were obtained from the Bosphorous database and comparisons were made to assess the correct matching ability between two faces. The R2 values derived from regression are used to quantify the degree of similarity between two faces. The results show that using tridimensional regression significantly increases the ability to correctly match two facial images that are represented by landmark data. Furthermore, use of the Euclidean transformation over affine and projective transformation presents better match results.