Automatic histogram-based fuzzy C-means clustering for remote sensing imagery

Ghaffarian S., Ghaffarian S.

ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING, vol.97, pp.46-57, 2014 (SCI-Expanded) identifier identifier


Fuzzy C-means (FCM) clustering has been widely used in analyzing and understanding remote sensing images. However, the conventional FCM algorithm is sensitive to initialization, and it requires estimations from expert users to determine the number of clusters. To overcome the limitations of the FCM algorithm, an automatic histogram-based fuzzy C-means (AHFCM) algorithm is presented in this paper. Our proposed algorithm has two primary steps: 1 - clustering each band of a multispectral image by calculating the slope for each point of the histogram, in two directions, and executing the FCM clustering algorithm based on specific rules, and 2 - automatic fusion of labeled images is used to initialize and determine the number of clusters in the FCM algorithm for automatic multispectral image clustering. The performance of our proposed algorithm is first tested on clustering a very high resolution aerial image for various numbers of clusters and, next, on clustering two very high resolution aerial images, a high resolution Worldview2 satellite image, a Landsat8 satellite image and an EO-1 hyperspectral image, for a constant number of clusters. The superiority of the new method is demonstrated by comparing it with the well-known methods of FCM, K-means, fast global FCM (FGFCM) and kernelized fast global FCM (KFGFCM) clustering algorithms, both quantitatively by calculating the DB, XB and SC indices and qualitatively by visualizing the cluster results. (C) 2014 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved.