Abstract: A texture image fuzzy segmentation algorithm based on dual-tree complex wavelet transform is studied in this paper. Texture features of an image are extracted using dual-tree complex wavelet transform. The fuzzy C-means clustering algorithm is directly applied to the texture segmentation, but traditional membership function in fuzzy C-means clustering algorithm is designed based on the distance between a sample and its cluster center, which is irrational for dataset with non-spherical-shape distribution. So, the fuzzy connectedness among samples is introduced to modify the traditional membership function. Simulations are performed on the presented algorithm, and the simulation result shows that the presented algorithm not only has high accuracy of boundary locations but also has good region homogeneity.

Key words: texture segmentation, feature extraction, dual-tree complex wavelet transform, fuzzy C-means clustering, membership function

摘要： 提出了一种基于对偶树复小波变换的模糊纹理图像分割算法，该方法包括纹理特征提取和纹理分类两个阶段，其中，特征提取在对偶树复小波变换的基础上进行；纹理分类可以直接用模糊C均值算法进行聚类从而完成纹理的分割，但由于该算法中隶属度函数是基于样本到类中心的距离设计的，这对非球形分布数据很不合理，针对该问题，引入样本与样本的紧致度来度量类中各个样本之间的关系从而修正隶属度函数，并将其用于纹理分类。实验结果表明与模糊C均值算法在运行时间上相差不大的情况下，改进的方法在分割精度、边缘准确性和区域一致性上都得到了明显的改善。

关键词: 纹理分割, 特征提取, 对偶树复小波变换, 模糊C均值聚类, 隶属度函数