基于形态学特征的颗粒图像分割和计数

Segmentation and Counting of Granular Objects Based on Morphological Features

针对黏连严重、分辨率和清晰度较低的颗粒图像分割和计数问题,提出了一种基于分水岭和形态学特征的新方法.首先使用分水岭算法分割图像,得到过分割的结果,接着通过本文定义的区域形态学特征,再根据加权马氏距离和区域连接图(Region Adjacency Graph)进行区域合并.采用实际图像进行的实验表明,该算法效果良好,对原始图像的灰度、对比度和噪声变化具有不变性,在准确率-查全率(Precision-recall)曲线的表现上优于现有方法.

For the task of segmentation and counting of granular objects with low quality images, a new meth- od which based on watershed and morphology is presented. First, an oversegmented result is gotten by conduc- ting the watershed algorithm to the smoothed image. Then the region merge stage, which is directed by using the morphological features defined in this article, as well as weighted Mahalanobis distance and region adja- cency graph （RAG） , takes place. Experiments with real images demonstrate the validity of the method, that it is invariant under translation of gray scale, contrast and noise. And it performs better than any present method, in terms of precision-recall criterion.