Nonlinear Discriminative Common Vector Method

In this paper we propose a new method called the Kernel Discriminative Common Vector (Kernel DCV) method. Firstly the original input space is mapped nonlinearly to a higherdimensional feature space through a kernel mapping. Then, the linear Discriminative Common Vector (DCV) method is applied in the transformed space. The proposed method employs the projection vectors from the null space of the within-class scatter matrix of the transformed samples for feature extraction. The same discriminative common vector for all samples in each class is obtained after feature extraction. Therefore, a 100% recognition rate is always guaranteed for the training set samples. The experiments on the test sets also show that the generalization ability of the proposed method compares favorably with the other kernel approaches. Also the fact that the test sample feature vectors are compared to only the discriminative common vectors, as opposed to all training set sample feature vectors, makes the proposed method ideal for real-time applications.