基于空间约束模糊聚类的活性污泥显微图像分割  

Segmentation of Activated Sludge Microscopic Images Using Spatially Constrained Fuzzy Clustering

作　　者：赵立杰 张晶 郭烁 黄明忠 王国刚 ZHAO Lijie;ZHANG Jing;GUO Shuo;HUANG Mingzhong;WANG Guogang(School of Information Engineering,Shenyang University of Chemical Technology,Shenyang 110142,China)

机构地区：[1]沈阳化工大学信息工程学院,辽宁沈阳110142

出　　处：《沈阳大学学报（自然科学版）》2023年第2期123-131,F0003,共10页Journal of Shenyang University：Natural Science

基　　金：国家重点研发项目计划(2018YFB1700200);2020年辽宁省高等学校创新人才支持计划;2021年度高等学校基本科研项目(重点项目)(LJKZ0442)。

摘　　要：针对活性污泥相差显微图像中絮体和丝状菌分布不均衡、絮体结构多样、存在大量细小丝状菌、图像对比度低,以及光晕、阴影等现象,提出了一种基于多尺度Retinex和空间约束模糊聚类的丝状菌分割算法和基于阈值、形态学操作的絮体分割算法。丝状菌分割算法对原始图像进行多尺度Retinex增强改善不同样本的光源强度,顶帽、底帽滤波增强絮体和丝状菌的对比度,采用边缘检测方法去除图像背景,考虑显微图像相邻像素空间的邻域信息采用空间约束模糊聚类法对增强后的图像进行聚类,结合像素面积和回旋半径形态学特性参数确定丝状菌目标和去除碎片;絮体分割算法利用灰度变换、图像反转以及开操作对絮体进行预处理,克服图像中的光晕和阴影,用阈值分割算法分割絮体,利用形态学操作对其进行后处理。实验结果表明,所提算法在图像分割准确率、精确率、Dice等指标上有明显提升。Aiming at the uneven distribution of flocs and filamentous bacteria in activated sludge phase contrast microscopic images,diverse floc structures,the presence of a large number of fine filamentous bacteria,low image contrast,halo and shadow,etc.,a filamentous segmentation algorithm based on multi-scale Retinex and spatially constrained fuzzy clustering and a floc segmentation algorithm based on threshold and morphological operation were proposed.The filamentous segmentation algorithm performs multi-scale Retinex enhancement on the original image to improve the light source intensity of different samples,and uses top cap and bottom cap filtering to enhance the contrast between floc and filamentous bacteria;the edge detection method is used to remove the image background,considering the spatial neighborhood information of adjacent pixels of microscopic images,and the spatial constraint fuzzy clustering method is used to cluster the enhanced images;pixel area and cyclotron radius morphological parameters are combined to determine filamentous bacteria targets and remove debris.The floc segmentation algorithm uses grayscale transformation,image inversion and opening operation to preprocess the floc to overcome the halo and shadow in the image;the threshold segmentation algorithm is used to split the floc;the image is post-processed using morphological manipulation.Experimental results show that the proposed algorithm has significantly improved the accuracy,precision,Dice and other indicators of image segmentation.

关 键 词：活性污泥 显微图像 图像分割 多尺度RETINEX 空间约束模糊聚类 OTSU分割 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]