一种混合特征高效融合的视网膜血管分割方法

发布时间：2018-05-12 13:18

本文选题：机器学习 + 视网膜　；参考：《电子与信息学报》2017年08期

【摘要】：将机器学习运用到视网膜血管分割当中已成为一种趋势,然而选取什么特征作为血管与非血管的特征仍为众所思考的问题。该文利用将血管像素与非血管像素看作二分类的原理,提出一种混合的5D特征作为血管像素与非血管像素的表达,从而能够简单快速地将视网膜血管从背景中分割开来。其中5D特征向量包括CLAHE(Contrast Limited Adaptive Histgram Equalization),高斯匹配滤波,Hesse矩阵变换,形态学底帽变换,B-COSFIRE(Bar-selective Combination Of Shifted FIlter REsponses),通过将融合特征输入SVM(支持向量机)分类器训练得到所需的模型。通过在DRIVE和STARE数据库进行实验分析,利用Se,Sp,Acc,Ppv,Npv,F1-measure等常规评价指标来检测分割效果,其中平均准确率分别达到0.9573和0.9575,结果显示该融合方法比单独使用B-COSFIRE或者其他目前所提出的融合特征方法更准确有效。
[Abstract]:It has become a trend to apply machine learning to retinal vascular segmentation. However, the selection of features as the characteristics of blood vessels and non-vessels is still a question of concern. Based on the principle of treating vascular pixels and non-vascular pixels as two categories, this paper presents a hybrid 5D feature as the expression of vascular pixels and non-vascular pixels, so that the retinal vessels can be separated from the background easily and quickly. Among them, 5D feature vectors include CLAHE(Contrast Limited Adaptive Histgram equalization, Gao Si matched filter and Hesse matrix transformation, morphological bottom hat transform, B-COSFIREB Bar-selective Combination Of Shifted FIlter reconstruction, and the needed models are obtained by training the fusion features into SVM (support vector machine) classifier. Based on the experimental analysis in DRIVE and STARE database, the segmentation effect is detected by using conventional evaluation indexes, such as DRIVE and STARE. The average accuracy is 0.9573 and 0.9575 respectively. The results show that the proposed fusion method is more accurate and effective than using B-COSFIRE alone or other existing fusion feature methods.
【作者单位】：北京工业大学信息学部;
【基金】：国家自然科学基金(61201360)~~
【分类号】：R770.4;TP391.41

【参考文献】