DCT子空间的邻域加权模糊C均值聚类算法

摘要

Fuzzy c-means clustering is an effective method used in image segmentation, but it is corrupted by noise easily because of ignoring spatial contextual information and structure information.A neighbourhood weighted fuzzy c-means clustering method based on DCT subspace is proposed. This papper first applies the discrete cosine transform (DCT) on image patches combined with the idea of partitioning, it establishes a similarity measure model based on image pacthes and local information. Then defines the neighbourhood-weighted distance to replace the Euclidean distance in the objective function. Finally, applied this method to synthetic image with different noises, real-world images, as well as magnetic resonance images. The experimental results show that the proposed algorithm can obtain more precise segmentation results and has the stronger anti-noise property.%模糊C均值聚类是一种有效的图像分割方法,但存在因忽略空间上下文信息和结构信息而易为噪声所干扰的现象。为此提出了DCT子空间的邻域加权模糊C均值聚类方法。该方法首先结合分块的思想,对图像块进行离散余弦变换(discrete cosine transform,DCT),建立了一个基于图像块局部信息的相似性度量模型；然后定义目标函数中的欧式距离为邻域加权距离；最后将该方法应用于加噪的人工合成图像、自然图像和MR图像。实验结果表明,该方法能够获得较好的分割效果,同时具有较强的抗噪性。