摘要： 图割法因无法体现像素点的纹理区域特性而难以应用于纹理分割。针对该问题，提出一种基于滤波器阵列与图割的彩色纹理分割算法。利用构建的滤波器阵列提取图像的纹理特征，并加入图像的H、S、I分量值组成纹理-色彩特征向量，采用texton直方图作为彩色纹理的统计模型对纹理-色彩特征向量进行统计计算，通过直方图差计算像素点间的纹理相似度，再应用图割法中的规范割准则对彩色纹理进行分割。实验结果证明，该算法具有较高的分割准确性。

关键词: 彩色纹理, 图像分割, texton直方图, 图割, 规范割

Abstract: It is difficult to apply graph cut to texture segmentation because it can not reflect the regional characteristics of the texture pixel. Aiming at the problem, a color texture image segmentation approach based on filter bank and graph cut is presented. Texture features of image which are extracted by the filter bank is connected with H, S and I components of the image to compose texture color feature vector. The color texture is modeled with texton histograms to statistic texture color feature vector, and the pairwise texture similarity is computed by comparing windowed texton histograms. Normalized cut criterion of graph cut method is applied to solve the optimal segmentation problem with the texton histograms. Efficiency and accuracy of the method are demonstrated by the experimental results.

Key words: color texture, image segmentation, texton histogram, graph cut, normalized cut

中图分类号: 

• TP317.4