分水岭算法分割显微图像中重叠细胞

Application of Watershed Algorithm for Segmenting Overlapping Cells in Microscopic Image

为实现医学临床显微图像自动快速分析,通过先将二值化后的图像进行距离变换,然后采用快速灰度重建算法重建距离变换后图像,最终用分水岭算法分割变换图像,有效地避免了为防止过分割而提取分水岭标记点过分依赖于图像先验知识的缺陷,实现自动探测目标细胞并分割重叠细胞,并使其适合于临床对算法速度的要求。将算法进行了C++程序实现,并应用于实际临床脱落细胞和病理免疫组化显微图像的自动分割。经过多幅不同疾病、不同背景的临床图像的分割验证,在光照均匀的情况下,该算法可以快速实现图像中细胞的提取以及粘连细胞的自动分割,完成一幅768×576图片的分割在AMD1600+的CPU上处理时间小于2 s,分割效果得到主任医师的认可,因此,该算法应用于临床细胞图像的分割是可行的。

In order to realize automatic and fast analysis of microscopic image, in this paper, the binarized image was first transformed to distance format. Fast grayscale reconstruction algorithm was then employed to reconstruct the transformed image. Finally Watershed algorithm was used to segment the result image. In this way, the defect of the watershed algorithm that the watershed marker extracting often depends on prior knowledge of the image is avoid and the algorithm to be suitable for real clinic practice and the auto segmentation of conglutination cells is realized. At last the algorithm was programmed with C ＋＋ and was used to segment the real clinical fall off cell and pathology cell images. After being applied to many different disease and different background cell image, result shows that the algorithm can pick up destination cells in a given image and automatically segment overlap cells accurately and rapidly under the condition of even illumination. For a given 768 × 576 image, the algorithm can finish the segmentation within 2 seconds in a AMD 1600 ＋ CPU. And the segmentation result is certificated by director doctor. Thus it is feasible to apply the algorithm to the segmentation of clinical cell image.