|
Title:
|
SPATIAL-SPECTRAL BASED HYPERSPECTRAL IMAGE CLUSTERING AN ADAPTIVE APPROACH USING CLUSTER'S BANDS BOX-PLOTS |
|
Author(s):
|
Mohamed A. Al Moghalis, Osman M. Hegazy, Ibrahim F. Imam, Ali H. El-Bastawessy |
|
ISBN:
|
978-989-8533-22-7 |
|
Editors:
|
Katherine Blashki and Yingcai Xiao |
|
Year:
|
2014 |
|
Edition:
|
Single |
|
Keywords:
|
Hyperspectral Images, clustering, image processing, k-means, remote sensing |
|
Type:
|
Short Paper |
|
First Page:
|
343 |
|
Last Page:
|
347 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Remote sensing Hyperspectral Image (HSI) has attracted researchers as a rich source of information. A recent trend in HSI analysis directs research to combine features of spatial nature with spectral nature. The motivation was that contiguous pixels share spectral features due to the low spatial resolution. This paper introduces an incremental work to a proposed approach that clusters HSI using K-means and combines spatial and spectral features together. The proposed approach uses selects pixels neighborhood, named kernel, from HSI to represent a cluster. For each cluster, pixels' bands box-plots profile is built to represent cluster bands distribution. Hence, the profile interprets the spectral features of a cluster. The approach starts by selecting K kernels from the given image scene randomly. However, results are affected by this random selection. In this paper, kernels are selected randomly but, they will be subjected to a test that avoids negative effect of randomness and prevents clusters to be represented again. In other words, the test ensures that successive kernels are different than previous selected kernels. Successive kernels are mapped against previous kernels box-plots profiles. Only kernels that prove to be spectrally different are used to represent consecutive clusters centroids. A pixel will join the cluster which score the minimum number of outlier reflectance values. |
|
|
|
|
|
|
