Abstract:

The key technique for solving the problem of kernel principle component analysis is to determine the several biggest eigenpairs of the kernel matrix. Lanczos algorithm is applied to the problem. Basing on the method, along with the thick-restart strategy, a highly effective algorithm is laid out. A corresponding visual C++ program for computing the parameters of a kernel component analysis model on large scale samples is developed on clapack and nu-TRLan software packages. Experiments are conducted to process hyperspectral remote sensing images to get the kernel principle components. The results illustrate the practicability of the algorithm in computing the parameters of a kernel principle component analysis model for large scale samples.

Key words: large scale samples, kernel principle component analysis, Lanczos algorithm, Thick-restart strategy