<title>K-means reclustering: algorithmic options with quantifiable performance comparisons</title>

Abstract

This paper presents various architectural options for implementing a K-Means Re-Clustering algorithm suitable for unsupervised segmentation of hyperspectral images. Performance metrics are developed based upon quantitative comparisons of convergence rates and segmentation quality. A methodology for making these comparisons is developed and used to establish K values that produce the best segmentations with minimal processing requirements. Convergence rates depend on the initial choice of cluster centers. Consequently, this same methodology may be used to evaluate the effectiveness of different initialization techniques. 1.Pixel classifiers, originally intended for multi-band images, assign individual pixels to specific classes based on their spectral properties. Pixel classifiers are furthered identified as supervised, unsupervised, or a

<title>Detecting human settlements in satellite images</title>

“…In future work, we hope to do such timing studies, and to explore both supervised learning techniques, especially Decision Tree and Artificial Neural Network classifiers, for various stages of the processing, as well as to examine unsupervised techniques such as isodata 16 and k-means re-clustering. 17,18 We also hope to establish robustness to geo-cultural, seasonal, illumination, and look-angle variations. We anticipate that several parameter sets will be required to produce global settlement maps, with each parameter set optimized for a particular portion of a continent and season.…”

Abstracting GIS layers from hyperspectral imagery

“…The most notable are determining the number of clusters K and spatially connecting the results. K-Means clustering is used to spectrally cluster pixels in [4], where K was varied from 2-16 yielding 101-695 separate spatial regions. Connecting these regions into meaningful abstractions that are separately labeled is an insurmountable task.…”

Unsupervised Segmentation of Hyperspectral Images Using Modified Phase Correlation

“…Unsupervised hyperspectral image segmentation using weighted incremental neural-network-based neuro-fuzzy systems has been proposed in [9]. The K-means algorithm is a well-established unsupervised method for image segmentation, and the utilization of the K-means reclustering algorithm for hyperspectral image segmentation is presented in [10]. Hyperspectral image segmentation using a multicomponent hidden Markov chain model has been proposed in [11].…”