Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 496 - Dimension Reduction
Type: Contributed
Date/Time: Thursday, August 6, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Statistical Learning and Data Science
Abstract #313315
Title: Canonical Correlation Analysis and Fusion Methods on a Large Face Database for Computer Vision
Author(s): Cuixian Chen* and Jasmine Gaston and Summerlin Thompson and Suhaela Eledkawi and Caroline Werther and Yaw Chang and Yishi Wang and Guodong Guo
Companies: University of North Carolina, Wilmington and University of North Carolina Wilmington and University of North Carolina Wilmington and Wright State University and University of North Carolina Wilmington and University of North Carolina Wilmington and University of North Carolina Wilmington and West Virginia University
Keywords: Canonical Correlation Analysis (CCA); Regularized CCA (rCCA); Kernel CCA (KCCA); Feature Fusion; Principal Component Analysis; computer vision
Abstract:

In 2014, Guo and Mu introduced a framework to simultaneously estimate age, gender, and race using canonical correlation analysis (CCA) based methods for dimension reduction, and multivariate regression techniques on a large face database of MORPH-II. This study expands their framework by considering additional facial feature extraction methods, as well as, Feature Fusion, a strategy that involves integrating multiple features to improve classification performance. The efficacy of the CCA based methods are compared using individual features and combinations of the best feature subsets. For fusion methods, due to the large dimensions of the fused feature data sets, Principal Component Analysis and 2-Stage CCA/KCCA methods are further considered in this study. The results indicate that overall, kernel CCA methods perform better than linear CCA methods, while being much more computationally intensive. Moreover, the results shows that the fusion based methods outperforms any of three individual features that are considered in this study.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program