融合Otsu与PCNN的细胞显微图像分割  

Cell Microscopic Image Segmentation Based onOtsu and PCNN

作　　者：胡蓉 夏平[1,2] 雷帮军[1,2] 徐光柱[1,2] 邹耀斌[1,2] Hu Rong;Xia Ping;Lei Bang-jun;Xv Guang-zhu;Zou Yaobin(Hubei Key Laboratory of Intelligent Vision based Monitoring for Hydroelectric Engineering,Three Gorges University,YiChang 443002,China;College of Computerand Information Technology,Three Gorges University,YiChang 443002,China)

机构地区：[1]三峡大学水电工程智能视觉监测湖北省重点实验室 [2]三峡大学计算机与信息学院,湖北宜昌443002

出　　处：《信息通信》2018年第11期15-19,共5页Information & Communications

基　　金：湖北省创新群体项目(2015CFA025);国家重点研发计划资助(2016YFB0800403);国家自然科学基金(联合基金)项目(U1401252);三峡大学培优基金项目(2018SSPY089)

摘　　要：针对细胞显微图像分割中由于纹理复杂较难定位轮廓信息的问题,提出了融合最大类间方差(Otsu)与脉冲耦合神经网络(Pulse Coupled Neural Network,PCNN)的细胞图像分割算法。为定义PCNN中初始阈值,应用Otsu算法,确定细胞图像中目标与背景间最大方差,此为统计意义上的最佳阈值,以此作为PCNN细胞分割中初始阈值,提高了PCNN网络收敛速度;其次,构建PCNN细胞分割网络,通过初始化网络参数,实现细胞图像迭代分割;从视觉效果和客观评价指标两方面验证表明,相对于经典的最大类间方差(Otsu)算法和脉冲耦合神经网络(PCNN)算法,采用该算法分割的细胞显微图像的效果,其边缘与细节信息的清晰度、精细度均有较大程度改善。For the problem that it＇ s difficult to locate contour information due to complex texture in segmentation of cell microscopic images, a cell image segmentation algorithm is proposed that combines the maximum between-cluster variance （Otsu） and pulse coupled neural network （PCNN）.In order to define the initial threshold in PCNN, Otsu algorithm is applicated, which determines the maximum variance between the target and the background in the cell image. This is the optimal threshold in the statistical sense. This is used as the initial threshold in PCNN ceU segmentation, which improves the convergence speed of the PCNN network.Then, building the PCNN cell segmentation network. And the iterative segmentation of cell images is achieved by initializing the network parameters. Visual effects and objective evaluation indicators shows that, compared to the classical Otsu algorithm and pulse coupled neural network （PCNN） algorithm, the effect of the cell microscopic image segmentation of the proposed algorithm is better. Its edges ,the clarity and finesse of details information have been greatly improved.

关 键 词：图像分割 细胞显微图像 脉冲耦合神经网络 最大类间方差法 阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]