一种综合的统计纹理分割方法

A synthetic statistic texture segmentation method

本文提出了一种无需任何先验知识的纹理图像分割系统。该系统采用模板滤波来提取纹理特征值，并用动态聚类法估计各类纹理的统计参数；然后引入空间约束条件对初始概率分布函数进行修正，并根据修正后的概率分布函数来确定各个象素所属的纹理类别。实验证明该系统无论是对人工纹理还是自然景物。

A texture image segmentaion system needing no prior knowledge was researched in this paper. The system used mask filtering for textures features, and estimated the statistical parameters of different textures with clustering algorithm. The spatial constrains were enforced to correct the probabilistic distribution functions, and the segmentation was carried out according to the resulting functions. The experiments showed the system has good results in either artificial textures or natural images.