结合显著性与Otsu法的紫花苜蓿茎显微图像分割  

Microimage segmentation combining salience with Otsu method for Medicago sativa stems

作　　者：金额尔都木吐 王春光[1] 李斯琴高娃 杜海峰 JIN Eerdumutu;WANG Chunguang;LI Siqingaowa;DU Haifeng(College of Mechanical and Electrical Engineering,Inner Mongolia Agricultural University,Hohhot 010018,China)

机构地区：[1]内蒙古农业大学机电工程学院,呼和浩特010018

出　　处：《黑龙江畜牧兽医》2021年第14期89-93,97,155,156,共8页Heilongjiang Animal Science And veterinary Medicine

基　　金：国家自然科学基金项目(32060415)。

摘　　要：为了解决紫花苜蓿茎显微结构形状各异、尺寸过小导致准确分割目标图像十分困难等问题,试验选择孕蕾期紫花苜蓿茎小段组织制作茎横切面石蜡切片,并对组织细胞进行染色;以图像显著性与最大类间方差法(Otsu法)为基础,对紫花苜蓿茎横切面、维管束木质化细胞、髓薄壁细胞图像进行预分割处理;运用数学形态学处理连接邻近物体、平滑边界及去除残留杂点;通过图像像素指标对茎横切面、维管束木质化细胞和髓薄壁细胞图像分割算法进行评价。结果表明:运用的数学形态学空洞填充和连通域标记法对茎横切面图像处理效果显著,最大连通区域即目标区域;利用数学形态学开运算法对维管束木质化细胞二值图像剔除小区域部分,其分割效果较理想;采用数学形态学孔洞填充和开运算法处理髓薄壁细胞图像,有效剔除了残留杂点,且目标区域分割较完整。紫花苜蓿茎横切面、维管束木质化细胞和髓薄壁细胞图像分割算法平均分割误差分别为2.537%、9.989%、6.197%。说明显著性与Otsu法结合算法可将对紫花苜蓿茎石蜡切片、染色、采集图像等环节导致的低质量图像实现准确分割,提取的前景目标形态完整、轮廓清晰。In order to solve the problems of the different shapes of the stems of Medicago sativa and the difficulty of accurately segmenting of target image caused by too small size,small sections of the stems of Medicago sativa during the gestation period were selected to make paraffin sections of the stems and tissue cells were stained.Based on the image saliency and Otsu method,and pre-segmentation was performed on the images of Medicago sativa stem cross-section,vascular bundle lignified cells,and medullary parenchyma cells.Mathematical morphology was used to connect adjacent objects,smooth borders and remove residual noise.The image segmentation algorithm of stem cross-section,vascular bundle lignified cells and medullary parenchyma cells were evaluated by image pixels index.The results showed that the mathematics morphology hole filling and connected domain labeling method had a significant effect on the stem cross-section image processing,and the largest connected area was the target area.The mathematical morphology open operation method was used to eliminate small areas from the binary image of vascular bundle lignified cells,and the segmentation effect was better.The mathematics morphology hole filling and opening operation were used to process the medullary parenchyma cell image,which effectively removed the residual noise,and the target area segmentation was relatively complete.The average segmentation errors of Medicago sativa stem cross-section,vascular bundle lignified cells and medullary parenchyma cells were 2.537%,9.989%,and 6.197%,respectively.It indicated that the new structure algorithm in this paper could achieve accurate segmentation for the low quality images caused by Medicago sativa stem paraffin sectioning,staining,image acquisition and other links,and the extracted foreground objects had complete shapes and clear outlines.

关 键 词：紫花苜蓿 茎解剖结构 石蜡切片 显微图像 图像分割 最大类间方差法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]