Application of improved fuzzy c-means algorithm to texture image segmentation

This study presents an efficient method for texture image segmentation based on dual-tree complex wavelet transform (DT-CWT)and improved Fuzzy C-means (FCM) algorithm. The procedure toward complete segmentation consists of two steps: texture feature extraction and feature classification. Firstly, texture feature is extracted in dual-tree complex wavelet domain for its shift invariance and more direction selectivity, we choose mean and variance of six high-frequency magnitude sub-bands as the texture features. Secondly, the fuzzy c-means algorithm is applied to the feature classification, but due to the random selectivity of initial clustering center, the clustering seeds may be too close which makes the FCM algorithm easily fall into local minimum, aiming at the problem, a new method based on maximun distance is proposed. In addition, to improve the membership function, the fuzzy connectedness of samples in the same cluster is proposed. Compared with the FCM algorithm, the experimental results show that the presented algorithm is more effective in texture image segmentation. At the same time, the presented algorithm is well applied to the segmentation of aero-image corrupted by noise. © 2013 Asian Network for Scientific Information.