一种新的基于图论聚类的分割算法

New Segmentation Technique Based on Clustering of Graph Theory

针对传统图论聚类法在分割图像时对噪声和模糊边界敏感,产生伪割集以及计算复杂度大的问题,对传统算法进行了相应的改进,即首先将每个像素作为一类改为将图像中灰度相同的像素作为一类;其次在计算权值时改进权函数定义,将节点与区域间的空间近邻关系约束进权函数表达式,而非传统算法中仅考虑节点与节点间的灰度和位置关系。对比实验表明,该算法只需要设计少量的参数即可自动完成聚类,所需的存储空间以及实现的复杂度相比于传统图论聚类法都得到极大改善。

In image segmentation,the traditional method of graph theory clustering is sensitive to salt ＆ pepper noise and fuzzy edges, produces much pseudo cutting-set and computational complexity. So, an improved approach is proposed in the paper. First, a newly classifying method is proposed, which is based on the same gray values, rather than the commonly used each pixel. Then, the definition of weight function is improved, which considers the spatial relation between pixel and regions, not only the gray values and position relation between two pixels. Experimental results show that the new method can automatically segment images with fewer parameters to other contrast algorithms, and the required storage space and realized complexity get satisfied improvement to the traditional algorithm＇s.