Discriminative learning of multiset integrated canonical correlation analysis for feature fusion

Based on the consideration that multiset integrated canonical correlation analysis (MICCA) does not include the class information of the samples, this paper presents a discriminative learning version of MICCA, called discriminative-analysis of multiset integrated canonical correlations (DMICC). The extracted features by DMICC not only contain the class information of training samples, but also possess more powerful discriminant ability than those by MICCA. The proposed DMICC method is evaluated on the AR, ORL face image databases, and the COIL-20 object database. The experimental results on face and object recognition demonstrate that DMICC is significantly superior to MICCA.