摘要
目的提出一种基于加权像素距离和相对熵的模糊C均值(Fuzzy C-means,FCM)改进算法。方法第一步采用局部平均隶属度倒数构建加权像素距离函数,第二步采用相对熵调整像素隶属度函数,最后根据改进的FCM聚类算法分割包含噪声的仿真图像和临床实例图像。图像分割结果评价指标采用分割错误率、划分系数、划分熵和Xie-Ben系数,并与标准FCM算法、基于局部数据距离的FCM算法、基于局部隶属度信息的FCM算法等算法进行比较。结果定性分析显示,本研究FCM改进算法获得的分割图像噪点最少,图像清晰度和对比度最佳。定量评价显示,采用本研究提出的FCM改进算法获得的图像分割评价指标均优于其他三种FCM算法,其中仿真图像分割错误率为0.09%±0.03%,划分系数为(0.9986±0.0003),划分熵为(0.0024±0.0009),Xie-Ben系数(0.2114±0.0019)。结论联合使用加权像素距离和相对熵能有效降低图像噪声,提高分割精度,具有较高的临床应用价值。
Objective To proposed a novel fuzzy C-means clustering algorithm based on weighted pixel distance and relative entropy for image segmentation.Methods First,the pixel to cluster-center distance was weighted using the reciprocal of the local membership average.Second,the regulation term was formulated using the relative entropy divergence which measures the proximity between a pixel membership and the local average of this membership in the immediate neighborhood.Finally,improved FCM clustering algorithm was employed to segment synthetic and real-world images.Image segmentation results was measured by misclassified pixels ratio,partition coefficient,partition entropy and Xie-Ben coefficient,and compared with the standard FCM,a local data-based information FCM and a type of local membership information based FCM algorithms.Results Qualitative analysis showed that the noise level was lowest,and contrast and definition were best with the proposed method.Quantitative evaluation results showed that the measured indexes of proposed method outperformed other three FCM-based algorithms,in which the simulated images misclassified pixels ratio was 0.09%±0.03%,partition coefficient(0.9986±0.0003),partition entropy(0.0024±0.0009),and Xie-Ben coefficient(0.2114±0.0019).Conclusion The combination of weighted pixel distance and relative entropy can effectively reduce image noise and improve segmentation accuracy,which has high clinical application value.
作者
王薇
魏应敏
WANG Wei;WEI Yingmin(Department of Radiology,Nanjing First Hospital,Nanjing Medical University,Nanjing Jiangsu 210006,China)
出处
《中国医疗设备》
2020年第4期71-74,共4页
China Medical Devices
关键词
图像分割
模糊C均值
加权距离
相对熵
image segmentation
fuzzy c-means
weighted distance
relative entropy
