Learning Supervised Covariation Projection Through General Covariance

Canonical correlation analysis (CCA) is a classical yet powerful tool for learning two-view feature representation. But, most CCA variants are based on the conventional covariance measure, which makes them difficult to uncover the complicatedly nonlinear relationship between distinct \textit{features}. In this paper, we address the preceding problem, and propose two novel CCA approaches in a supervised manner by defining a general covariance metric. The proposed approaches not only consider the label information of training data, but also the nonlinear relationship between features rather than \textit{samples}, which leads to greater flexibility in practical applications. A series of experiments on five benchmark datasets demonstrate the effectiveness of our proposed methods in terms of classification accuracy.