Class-Incremental Kernel Discriminative Common Vectors

In this paper, we propose an efficient algorithm for implementing the class-incremental kernel discriminative common vectors method via kernel method. One nonlinear discriminative common vector is computed for each class by projecting a sample in each class onto the orthonormal nonlinear discriminative vector. The orthogonalization procedure is performed twice in feature space which is only involved computing a kernel matrix and performing Cholesky decomposition on the kernel matrix. Thus, the real-time performance of classification is guaranteed. The theoretical justification is presented in this paper.