Initialization Methods For Conventional Fuzzy C-Means And Its Application Towards Colour Image Segmentation
Loading...
Date
2011-05
Authors
Tan, Khang Siang
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
Due to its capability in providing a particularly promising solution to clustering problems, the conventional Fuzzy C-Mean (FCM) algorithm is widely used as a segmentation method. However, it is very sensitive to the initialization conditions of number of clusters and initial cluster centres. Thus, three initialization schemes for the conventional FCM algorithm namely the Hierarchical Approach (HA), the Colour Quantization (CQ) and the Histogram Thresholding (HT) are proposed to automatically obtain the initialization conditions for the conventional FCM algorithm. Prior to the development of the initialization schemes, the Peak Finding Histogram Analysis (PFHA) algorithm is proposed to locate the modes and then the valleys between any two adjecent modes in the histogram. Then, the PFHA algorithm is applied to split the colour image into multiple homogeneous regions before employing the merging algorithm for the HA, the CQ and the HT initialization schemes in different ways. The experimental results show that the proposed initialization schemes outperform other conventional initialization schemes by reducing the classification errors and producing more homogeneous regions in the segmentation results.
Description
Keywords
Initialization methods for conventional fuzzy c-means , application towards colour image segmentation