高效的白细胞分割算法

白细胞分割是细胞形态学研究中一个重要、富有挑战性的课题。提出一种基于目标检测的白细胞分割算法。具体地讲，首先根据目标检测方法检测出白细胞，并由白细胞的位置信息得到包含白细胞的子图，然后运用多项式拟合的方法得到子图的灰度直方图的波谷值，再在子图上运用直方图阈值算法分割出细胞核。该算法在分割细胞核的过程中，既可以有效地避免血小板和红细胞等干扰，又能较容易地估计出阈值。对于白细胞的细胞质分割，将白细胞位置信息作为GrabCut算法中人工交互部分，通过迭代法分割出白细胞的细胞质。实验结果表明，该算法能准确地定位白细胞，并根据白细胞的位置信息可以降低白细胞分割的难度，提高其分割的精度和分割效率。特别地，在Cellavision的白细胞图片数据库的实验结果表明，所提的白细胞分割算法对不同类别、不同染色剂和不同拍摄环境下得到的白细胞都能得到较好的分割效果，同时算法又还具有很好的泛化性。

Efficient segmentation algorithm for white blood cell

White Blood Cell（WBC） segmentation is an important and challenging issue in the cell morphology. This paper proposes an efficient approach for WBC segmentation based on the method of objection detection. In detail, it first uses the method of object detection to locate WBC, and gets a sub-image containing the complete WBC according to this location information. Then, it obtains the threshold of sub-image’s gray histogram using the method of polynomial fitting, and segments WBC nuclear from the sub-image. This approach can effectively avoid the interference from the red blood cell and background, and it is easy to estimate the threshold. In the meanwhile, it segments WBC cytoplasm by means of treating the image window as the “trimap” in the GrabCut algorithm to minimize the manual intervention. Some experimental results show that the proposed algorithm can exactly detect WBC, and it can get a higher segmentation accuracy and better effect than some existing methods. Especially, in the experiments on the Cellavision dataset, it can not only effectively segment WBCs which is got from different types, dyes, and shoot situations, but also achieve a higher robustness.