基于非下采样Contourlet变换的视网膜分割  

Retinal Vessel Segmentation Using Nonsubsampled Contourlet Transform

作　　者：钟桦[1] 焦李成[1] 侯鹏[1] 

机构地区：[1]西安电子科技大学智能感知与图像理解教育部重点实验室,西安710071

出　　处：《计算机学报》2011年第3期574-582,共9页Chinese Journal of Computers

基　　金：国家自然科学基金(60970066;60972148);国家"九七三"重点基础研究发展规划项目基金(2006CB705707);北京市自然科学基金(7092020);教育部长江学者和创新团队支持计划(IRT0645);中央高校基本科研业务费专项资金(JY10000902032)资助~~

摘　　要：基于非下采样Contourlet变换(NSCT)良好的多尺度、多方向和平移不变性等优点,提出了一种基于NSCT的视网膜图像分割方法.该方法首先通过分析NSCT变换对血管的系数响应,提出基于NSCT的线状特征提取算法.随后对所提取的特征向量利用高斯混合模型(GMM)进行建模,并采用EM算法估计其参数.最后采用贝叶斯规则对血管和非血管像素进行分类,以达到图像分割的目的.在DRIVE和STARE两个数据库上的实验结果表明,基于NSCT的线状特征提取算法能够很好地表征血管目标,在血管的分割正确率和定位精度等指标上表现出良好的性能.A retinal vessel segmentation method based on Nonsubsampled Contourlet Transtorm （NSCT） is proposed. Due to the advantages of NSCT on multi-scale, multi-direction and translation invariance, it is very suitable for the representation of line singularity of retinal vessel. This paper analyzes in detail the NSCT coefficient responses for the retinal vessel and give a simple and pixel-wised feature extraction algorithm. Then the extracted feature vectors are modeled with Gaussian Mixed Model （GMM） whose parameters are estimated using Expectation-Maximization （EM） algorithm. Finally, Bayesian rules are used for the supervised classification of vessel pixels and nonvessel pixels. Because the Contourlet basis function can capture anisotropic geometrical structures efficiently by multi-direction selection, the NSCT coefficients of line object are sparser but more accurate than those represented by wavelet transformation. So the proposed algorithm is potential to achieve good performance either in segmentation accuracy and vessel location. The method＇s performance is evaluated on publicly available DRIVE and STARE databases. Experimental results show that the proposed method achieves good performance in segmentation accuracy when compared with state-of-the-art approaches. On the other hand, the method can obtain better precision in vessel location and the low rate of false detection, which can effectively avoid the conglutination of the parallel vessels.

关 键 词：视网膜分割 NSCT 线状特征 高斯混合模型 贝叶斯规则 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]