Supervised Independent Component Analysis by Maximizing J-Divergence Entropy

Independent component analysis (ICA) is one of the most exciting topics in the fields of neural computation, advanced statistics, and signal processing, which finds independent components from observing multidimensional data based on higher order statistics. The theory of independent component analysis is traditionally associated with the blind source separation (BSS). Since the recent increase of interest in ICA, it has been clear that this principle has a lot of other interesting applications, especially feature extraction. But traditional independent component analysis mainly aims at BSS and is not suitable for recognition and classification due to ignorance of the contribution of independent components to recognition performance. In order to overcome this problem, a new supervised ICA algorithm based on J-divergence entropy is proposed, which can measure the difference of classes. Experiment results of face and iris recognition show that the algorithm improves the performance efficiently.