中文摘要：

多层马氏链图像分割算法在纹理碎片化阶段基于颜色信息实现聚类，用于聚类的特征只包含颜色以及颜色之间的关系，导致最终的分割结果对颜色过度敏感。针对此问题提出基于纹理基元的纹理碎片化聚类方法，将颜色与灰度分布相结合构造为纹理基元，面向纹理基元的聚类分割使碎片结果稳定，避免了因光照等因素导致纹理的误分割。实验证明，改进后的算法分割精度有明显的提高，对于遥感图像分割来说，纹理分割结果的稳定性更高。

英文摘要：

The multiple Markov image segmentation algorithm realizes clustering in the texture fragmentation stage. That feature to cluster only includes color and relationship between colors leads to the final result overly sensitive to color. An algorithm is proposed based on primitive texture fragmentation clustering. Texture primitive is constructed based on color and gray level distribution. The clustering facing to the texture primitive is stable, avoiding error segmentation in the case of light etc. The experiment proves that the algorithm can improve in accuracy of the segmentation, to the re mote sensing image, the result is more stable.