自适应脉冲耦合神经网络与匹配滤波器相结合的视网膜血管分割

发布时间：2018-03-31 11:20

本文选题：视网膜图像处理　切入点：血管分割　出处：《光学精密工程》2017年03期

【摘要】：针对眼底图像中血管与背景间对比度低以及血管自身结构复杂等因素对视网膜血管分割所带来的问题,本文提出了一种具有自适应连接值的脉冲耦合神经网络(PCNN)与高斯匹配滤波器相结合的视网膜血管分割方法。首先,利用对比度受限制的自适应直方图均衡化(CLAHE)技术与二维高斯匹配滤波器对血管区域的对比度进行有效增强。然后,利用经验阈值选择出一定的血管区域作为初始种子区域。接着,将带有快速连接机制的PCNN与种子区域增长思想相结合,通过自适应动态设置PCNN中的连接强度系数和停止条件,实现眼底图像中血管区域的自动生长。整个算法在DRIVE视网膜图像库中进行了测试,实验结果表明,相比于不使用动态连接强度系数与停止条件的方法,所提出算法的灵敏度从49.79%提高至70.39%,准确度从95%提高到95.39%。证明了该算法具有较好的分割精确度和应用价值。
[Abstract]:Aiming at the problem of retinal vascular segmentation caused by low contrast between blood vessel and background in fundus image and complex structure of blood vessel itself, In this paper, a pulse coupled neural network with adaptive connection value (PCNN) and Gao Si matched filter are proposed to segment retinal blood vessels. The contrast-limited adaptive histogram equalization technique and two-dimensional Gao Si matched filter are used to effectively enhance the contrast of vascular region. A certain vascular region is selected as the initial seed area by using empirical threshold. Then, the PCNN with fast connection mechanism is combined with the seed region growth idea, and the connection strength coefficient and stopping condition in PCNN are set dynamically and adaptively. The whole algorithm is tested in the DRIVE retinal image library. The experimental results show that compared with the method without dynamic connection strength coefficient and stop condition, The sensitivity and accuracy of the proposed algorithm are improved from 49.79% to 70.39% and 95% to 95.39% respectively. It is proved that the algorithm has good segmentation accuracy and application value.
【作者单位】：三峡大学计算机与信息学院;湖北省水电工程智能视觉监测重点实验室(三峡大学);
【基金】：国家自然科学基金资助项目(No.61402259,No.U1401252) 湖北省自然科学基金创新群体计划项目(No.2015CFA025) 三峡大学2016年硕士学位论文培优基金资助项目(No.2016PY050)
【分类号】：R770.4;TP183;TP391.41

【相似文献】