摘要
图像分割是图像识别的关键,笔者曾经分别采用松弛迭代[1]和K均值聚类方法对胃上皮内肿瘤图像进行分割,实验表明这些算法对粘连严重的图像分割效果很差,故本文应用分水岭分割算法Vincent和Inver,对粘连情况不同的多类胃上皮内肿瘤图像进行了图像分割实验.实验结果表明:对于粘连较少的细胞图像,这两种算法都能较好地分离出目标细胞,但对于粘连严重的细胞图像,Inver算法的分割效果比Vincent要好,但Inver算法容易出现过分割现象.
Image segmentation is a key process of image recognition. Relaxation iteration algorithm and K-means cluster algorithm are used to resolve the stomach epidermis tumor segmentation, but some conglutinated cells cannot be separated by others. So the Vincent watershed algorithm and the Inver watershed algorithm are designed to conduct segmentation experiments about the stomach epidermis tumor. A lot of experiments about several kinds stomach epidermis tumor are done to build the segmentation theory. The conclusion of these experiments shows that these two algorithms could get good segmentation results when the cell image has little overlapping area. The Inver algorithm could get better result than the Vincent algorithm while the cell image has much more overlapping area. But Inver algorithm may have more useless segmentation areas than Vincent algorithm.
出处
《华东交通大学学报》
2009年第1期52-57,共6页
Journal of East China Jiaotong University
基金
江西省教育厅2008年科技计划项目(GJJ08237)
华东交通大学校立科研资助项目(O8XX06)
