【摘要】 为提高三维CT图像分割精度,提出先验知识增强的GrabCut分割方法。对一组事先手动精确分割的CT图像进行训练,利用主动形状模型(ASM)获得平均形状模型和形状变化量等统计形状先验知识;利用先验知识定义形状项,将目标的形状和位置信息融入GrabCut分割框架,增强待分割图像信息,约束能量函数获得全局最优解。对比实验结果表明,相比主动形状模型和传统GrabCut算法,该算法具有较高的分割精度,在训练集较小的情况下,能比主动形状模型获得更好的分割结果。更多还原

【Abstract】 To improve the accuracy of three-dimensional CT images segmentation,aprior knowledge enhanced GrabCut segmentation approach was introduced.A group of CT images that were manually segmented were trained in advance,and the active shape model(ASM)was used to obtain the statistical shape prior knowledge that included mean shape model and shape variance information.A priori knowledge was used to define shape terms,the shape and position information of the object was integrated into the GrabCut framework,which enhanced the information of the image to be segmented and made the energy function obtain the global optimal solution.Contrast experimental results show that the algorithm has higher segmentation accuracy compared with the traditional GrabCut and active shape models.In the case of smaller training set,it can get better segmentation results than active shape model.更多还原