Abstract
The fuzzy c-means (FCM) clustering algorithm has been widely used in image segmentation. However, FCM exhibits poor robustness to noise, often leading to unsatisfactory segmentations on noisy images. Additionally, the FCM algorithm is sensitive to the choice of initial cluster centers. In order to solve these problems, this paper proposes clone kernel spatial FCM (CKS_FCM), which improves segmentation performance in several ways. First, in CKS_FCM, an immune clone algorithm is used to generate the initial cluster centers, which helps prevent the algorithm from converging on local optima. Second, CKS_FCM improves the robustness to noise by incorporating spatial information into the objective function of FCM. Third, CKS_FCM uses a non-Euclidean distance based on a kernels metric, instead of the Euclidean distance conventionally used in FCM, to enhance the segmentation accuracy (SA). We present experimental results on both real and synthetic SAR images, which suggest that the proposed method can generate higher accuracy, and obtain more robustness to noise, as compared against six state-of-the-art methods from the literatures.
Original language | English |
---|---|
Pages (from-to) | 1 - 13 |
Journal | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
Volume | PP |
Issue number | 99 |
DOIs | |
Publication status | Published - 2016 |
Keywords
- Fuzzy C-means (FCM) cluster
- immune clone algorithm
- kernels metric
- spatial information