A new robust fuzzy c-means clustering method based on adaptive elastic distance

摘要

Smoothing by neighborhood information is an effective way for clustering methods to improve the robustness of image segmentation. But the usual smoothing will make some important details lost, especially when the learned cluster structure is improper. On the contrary, a well learned cluster structure is beneficial to maximize the effect of smoothing so that performance of image segmentation could be promoted. However, in most state-of-art fuzzy c-means(FCM) based clustering methods. Regularization functions are applied to change the interaction between each pair of points, and it usually shows in a monotonous trend, either increase or decrease. It results in a poor ability to recognize cluster structure. To solve this problem, we introduced an adaptive elastic distance based on membership, and proposed an elastic fuzzy c-means (EFCM). EFCM provides a sparser description for reliable points and a fuzzier description for marginal points of clusters, thus, the interpretability of reliable points is improved and the effect of marginal points of clusters is also highlighted. It means, EFCM has a better ability to recognize intrinsic cluster structure. Additionally, by combining EFCM with a smoothing method, a new robust fuzzy c-means clustering method based on adaptive elastic distance (ARFCM) for image segmentation was proposed. Taking advantage of the improved ability to recognize intrinsic cluster structure, ARFCM can make better use of neighborhood information in image segmentation. Experiment results show that, for all images polluted by different noises, ARFCM achieves better segmentation accuracy than other state-of-art methods. Furthermore, ARFCM can gain clearer texture and more homogeneous regions in real images.