Abstract: An unsupervised texture image segmentation algorithm based on wavelet transform and mean shift algorithm is studied.Firstly，the original image is decomposed in two levels using wavelet transform algorithm.Secondly，the mean shift algorithm is used together with the fuzzy c-means algorithm to divide the data into an appropriate number of clusters in the coarse scale.Thirdly，a peer group corresponding to a clustering center reconstructed from the one of the coarse scale is automatically achieved by use of a threshold function and the Fisher discrimination.And then a texture feature clustering effect is achieved.At last，simulations are performed on the presented algorithm，and the simulation result shows that the presented algorithm not only has high accuracy but also can solve the problems of giving the number of cluster in advance and of sensitivity to initial clustering center of the traditional clustering.

Key words: itexture segmentation, wavelet transform, feature extraction, mean shift, fuzzy c-means

摘要： 提出了一种基于小波变换和均值偏移的无监督纹理图像分割算法。首先用小波变换对图像进行二级小波分解，然后用均值偏移算法估计出粗尺度上对应的聚类数目，并结合模糊c均值算法进行聚类，在此基础上，用定义的阈值函数和Fisher判据确定出细尺度上每个初始聚类中心的一个同组，从而实现图像的由粗到细的分割。实验结果表明，在分割精度相差不大的情况下，该方法解决了传统聚类方法所存在的需要聚类数目和对初始聚类中心敏感问题。

关键词: 纹理分割, 小波变换, 特征提取, 均值偏移, 模糊c均值