SEM材料图像目标区域分割

由SEM获取的纤维材料图像,目标孔洞粘联明显、重叠严重,特征参数提取困难。本文首先利用提取目标特征凹点的方法,实现了目标区域分割,但在分割精度上存在着不足,且算法较为复杂。对此,采用了一种基于距离变换的分水岭算法,通过对不同区域设置标记,循环标记的最终结果形成分水岭,并通过抑制过分割实现了对目标区域更为精确的分割。实验结果表明该算法具有良好的特性。

Region Segmentation for SEM Image Based on Watershed Transform and Concave Spot

The fibrous material image captured from Scanning Electronic Microscopy(SEM) has low SNR,and the objects in image overlap together.Region segmentation using concave spot extraction was firstly performed in this paper.But the results show that this method has shortages in precision,especially for complex overlapping segmentation.To solve this problem,the method based on watershed transform was further proposed in this paper.Distance transform was used for preprocessing the image to get gray level distribution.The watershed line was achieved to mark the boundaries of different objects.Experimental results show that the segmentation approach performs more successfully.