Main Article Content
The main purpose of k-Means clustering is partitioning patterns into various homogeneous clusters by minimizing cluster errors, but the modified solution of k-Means can be recovered with the guidance of Principal Component Analysis (PCA). In this paper, the linear Kernel PCA guides k-Means procedure using filter to modify images in situations where some parts are missing by k-Means classification. The proposed method consists of three steps: 1) transformation of the color space and using PCA to solve the eigenvalue problem pertaining to the covariance matrices of satellite image; 2) feature extraction from selected eigenvectors and are rearranged by applying the training map to extract the useful information as a set of new orthogonal variables called principal components; and 3) classification of the images based on the extracted features using k-Means clustering. The quantitative results obtained using the proposed method were compared with k-Means and k-Means PCA techniques in terms of accuracy in extraction. The contribution of this approach is the modification of PCA selection to achieve more accurate extraction of the water-body segmentation in satellite images.