Kernel projection classifiers with suppressing features of other classes
Neural Computation
Feature extraction using constrained approximation and suppression
IEEE Transactions on Neural Networks
| Hi-index | 0.00 |
We propose a new kernel-based method for pattern recognition. Support vector machine (SVM), principal component analysis (PCA), and Fisher discriminant have been extended to kernel based methods and they achieve better performance. In this paper, we propose kernel sample space projection classifier (KSP) for pattern recognition. In KSP, an unknown input pattern is discriminated by comparing the norms onto kernel sample spaces which are spanned by sample vectors mapped to a high dimensional feature space by Mercer kernel function. In this paper, we provide a closed form of our method and show its advantages by experimental results of the recognition problem using handwritten digit database "MNIST" and some two-class classification problems. Finally we compare it with other methods from several points of view.