Supervised Canonical Correlation Analysis and Its Application to Information Fusion

Based on the traditional canonical correlation analysis (CCA), by defining within-class correlation and within-class correlation matrix, a new supervised canonical correlation analysis (SCCA) method of multi-mode feature extraction is proposed. The theory of SCCA can be explained as follows: if a pattern has a pair of observations (For any pattern space, there are two observation spaces), SCCA can find a enhanced relevant subspace of the two observation spaces by adding the class label information, in which the mappings of the pair observations of the same patterns have maximum correlation, and the enhanced relevant subspace can have more semantic discriminate ability for pattern recognition. Our proposed algorithm is validated by the experiments on Yale face database. Compared with other methods, the recognition rate of our method is far higher than that of the PCA algorithm which only adopt single-mode features and the traditional multi-mode CCA feature fusion algorithm.