Decomposition Mixed Pixels of Remote Sensing Image Based on 2-DWT and Kernel ICA.

In this paper, we propose a novel method for decomposing mixed-pixels of remote sensing images, which integrates two-Dimensional Wavelet Transform (2-DWT) and Kernel Independent Component Analysis (KICA) technique. In order to improve the signal and noise ratio of the original mixed-pixel images, we apply wavelet analysis method to reduce the noise of the images. High-frequency sub-image in wavelet domain is approximately represented by a kind of super-Gaussian Laplace distribution, and KICA is adopted for this distribution with greater kurtosis for obtaining higher accuracy and faster convergence rate. The experiments show that decomposition result with the proposed method is much improved not only at accuracy but also remarkably robust to noise compared those obtained with 2-DWT-ICA or KICA. [ABSTRACT FROM AUTHOR]