Abstract: Aiming at CT image segmentation, an improved algorithm is proposed, which reduces the dependence of traditional knockout method on the manual input, and improves its robustness. Use the grayscale information of CT image to modify the original input. Add the processing of adaptive threshold and region grow to rectify the foreground, a negative gradient searching to correct the background, and the transformation of color space to get a new equation about alpha in the signal channel. Experimental results show that this improved model has a better behavior.

Key words: digital matting, CT image segmentation, alpha estimation, foreground, background

摘要： Knockout算法对初始交互轮廓线的依赖性较高、鲁棒性较差。为此，提出一种改进算法，引入CT图像的典型纹理特征信息对初始交互轮廓线进行修正，通过自适应阈值和区域生长法修正前景区域，利用负梯度搜索法修正背景区域，通过颜色空间变换将透明度的三通道表达式转换为单通道形式。实验结果表明，改进算法对初始交互轮廓线的依赖性较低，分割的准确度较高。

关键词: 数字抠图, CT图像分割, 透明度估计, 前景, 背景

CLC Number: 

• TP391.41