中文摘要：

针对大纹理图像分割困难的问题,提出一种大纹理图像分割算法.在获取影像的初始过分割区域后,依次使用区域颜色和背景对区域进行划分,得到区域标记图像;依据不同标记之间的空间交互强度,建立全局最优的标记合并序列,获取多粒度的分割结果;提出边界类别分布模型来建模区域或标记的空间交互关系.对比实验结果表明,该算法在处理大纹理图像分割方面有明显优势.

英文摘要：

To overcome the difficulties in segmenting macro-texture images, a novel segmentation algorithm was proposed. After obtaining over-segmented regions by the initial segmentation, these regions were split according to their spectral responses and contextual information in sequential order to get the region label image. Then, a global optimal label merging serial was built on the basis of the interactive intensity between different label pairs to obtain the multi-granular segmentation results. The boundary label distribution was proposed to model the spatial interaction between both regions and labels. The comparative experimental results verify the superiority of the proposed algorithm in segmenting macro-texture images.