一种基于邻域数组的区域合并新方法

A region merging method based on neighborhood array

针对图像分割中的过分割问题,该文提出了一种区域合并方法。该方法在区域合并时采用邻域数组的数据结构来维护区域间的邻接关系,该数据结构相对传统的邻接图具有更高的效率。为了能够综合利用分割对象的各种特征,兼顾区域合并的精度与效率,将合并过程分为两个步骤进行:第一步以区域面积阈值作为合并终止条件,在合并准则中只考虑灰度信息;第二步将灰度阈值作为合并终止条件,并在合并准则中加入了纹理信息。实验结果表明,该方法能够对初分割图像进行高效且精确的合并。

A region merging method was presented in this paper to solve the problem of over segmentation in image segmentation.The data structure of neighborhood array was adopted to maintain adjacencies between regions in the process of merging,and it was more efficient than the traditional one of the region adjacency graph.In order to utilize various features of the objects and take into account the accuracy and efficiency of region merging,the process was divided into two steps.Area threshold was taken as the termination condition of merging in the first step and only the gray information was considered in the merging rule.Gray threshold was taken as the termination condition of merging in the second step,and texture information was added into the merging rule.Experimental results showed that this method could make efficient and accurate merging result to early segmented image.