结合显著性的GrabCut及在骨髓细胞图像分割中的应用

针对传统GrabCut算法需要用户交互缺点,提出一种基于上下文感知显著性的GrabCut的改进的图像分割算法.首先用上下文感知得到待分割图像的显著图,然后由二值化的显著图确定GrabCut算法的初始化区域,再通过迭代使能量函数最小化分割出目标,算法应用于骨髓细胞图像分割上.实验结果表明,此算法能避免以往细胞分割算法如支持向量机、K-Means等参数调整问题,总体误差率较低,自动化程度高,鲁棒性强.

Combining saliency map and GrabCut for the segmentation of bone marrow cells images

In view of the disadvantages of user interaction in traditional GrabCut algorithms, an improved GrabCut algorithm for image segmentation based on the context-aware method was proposed. Firstly the saliency map for the bone marrow cells image was obtained by the context-aware algorithm. The initialization of GrabCut algorithm was determined by the binarized saliency map. And then the target was segmented by energy minimization. The algorithm was applied to bone marrow cell image segmentation. Experimental results show that it can avoid the parameters adjustment problem which arises in support vector machines and K-Means. It also has a relatively low overall error rate with high automation and strong robustness.