纹理图像分割过程一般分为特征抽取和特征划分．文中提出一种新的基于分形维数的纹理图像分割方法．在特征抽取上，以分形作为纹理特征，运用图像变换的思想，结合差分盒计数和基于分形布朗自相似模型的分形估计方法，对各种图像变换结果进行估计而得到基于分形维数的矢量；在特征划分上，再运用边缘保持的图像滤波对得到的分形维数特征空间进行滤波，滤波结果用Ｋ 均值簇分类法作特征空间的初分类；最后应用基于概率松驰标记方法对图像进行特征划分，完成纹理分割．实验表明此方法对综合和真实图像分割得到了很好的效果 Texture image segmentation process is generally divided into feature extraction and feature classification. In this paper, a new texture image segmentation method based on fractal dimension is proposed. In feature extraction, the fractal is taken as the texture feature, the idea of ​​image transformation is used, and the differential box counting and the fractal estimation method based on the fractal Brownian self-similarity model are used to estimate the results of various image transformations to obtain the vectors based on the fractal dimension. In feature classification, the edge-preserving image filtering is used to filter the feature space of the fractal dimension. The filtering result is classified by K-means clustering as the initial classification of the feature space. Finally, the image is characterized by using the probability relaxation marking method Divide, complete texture segmentation. Experiments show that this method has good effect on the synthesis and real image segmentation