Nonlinear Feature Extraction Using Generalized Canonical Correlation Analysis

Thomas Melzer, Michael Reiter, Horst Bischof

Abstract: This paper introduces a new non-linear feature extraction technique based on Canonical Correlation Analysis (CCA) with applications in regression and object recognition. The non-linear transformation of the input data is performed using kernel-methods. Although, in this respect, our approach is similar to other generalized linear methods like kernel-PCA, our method is especially well suited for relating two sets of measurements. The benefits of our method compared to standard feature extraction methods based on PCA will be illustrated with several experiments from the field of object recognition and pose estimation.