An implementation framework for kernel methods with high-dimensional patterns

As nonlinear feature extraction methods, kernel methods have been widely applied in pattern recognition. However, for high dimensional data such as face images, a kernel method will correspond to a high computational cost. In this paper, a novel idea and framework are presented to implement the kernel methods on high-dimensional data. A remarkable character of the framework is that there are two feature extraction processes. The first feature extraction process is performed to transform high dimensional samples into low dimensional data. And, the second feature extraction process is implemented based on the obtained low dimensional data. With the novel framework, the kernel methods will become much efficient. Moreover, all kernel methods can work with the framework. The experiments on face images show the validity of this framework. Further more, with this framework, kernel methods can achieve higher classification accuracies in comparison with the naive kernel methods.