摘要
水平集方法已经广泛应用于图像分割,Chunming Li等人早期的模型通过在能量方程中引入惩罚项可以避免重新初始化。但惩罚项中的函数会引起扩散率趋于无穷大的问题,因此Chunming Li等人通过改进惩罚项中的函数,解决了扩散率的问题。针对新模型采用高斯滤波去除图像噪声使图像边缘变模糊的问题,采用正则化的P-M方程滤波,去除噪声的同时保护图像边缘信息。同时,新模型仍然不能实现自适应分割。通过初始曲线内外梯度模值的信息改变曲线内法向量的方向,从而使曲线自适应地向内或者向外演化。最后,用改进的算法准确地提取出了医学图像的轮廓,算法的效率也有很大的提高。
Level set methods have been extensively used in image segmentation, Chunming Li et al. 's preliminary work add the penalty term to the energy function eliminates the need of the reinitialization produce. But the function of the penalty term can lead the diffusion rate go to infinity. So Chunming Li et al. improve the function of the penalty term to avoid this problem. This new model use Gaussian filter to reduce the image noise, but it can make blurred image edges. It can remove the noise while preserving edge information by adopts regularized P - M equation filter. The new model can't segment adaptively ,by using gradient information which depends on inward and outward area's curve to improve the direction of the inward normal vector, and the curve can adaptive evolution inward or outward. In the end, the modified algorithm is applied to accurately extract the medical image's contour, and greatly reduce computational cost.
出处
《微计算机应用》
2011年第12期19-23,共5页
Microcomputer Applications
基金
国家自然科学基金(编号61179011)
福建省自然科学基金(编号2010J01327)
关键词
水平集
正则化P-M方程
单位法向量
测地线活动轮廓
图像分割
Level set, Regularized P- M equation, Unit normal vector, Geodesic active contour, Image segmentation