基于变分水平集主动轮廓的叶片表皮细胞厚度测量方法研究  

Extraction and Thickness Measurement of Leaf Epidermal Cells Based on Active Contour of Variational Level Set Method

作　　者：孙小添 王海超 郭根胜[1] 裴志永[1] SUN Xiaotian;WANG Haichao;GUO Gensheng;PEI Zhiyong(College of Energy and Transportation Engineering,Inner Mongolia Agricultural University,Hohhot 010018,China)

机构地区：[1]内蒙古农业大学能源与交通工程学院,呼和浩特010018

出　　处：《森林工程》2022年第2期1-7,共7页Forest Engineering

基　　金：内蒙古自治区科技计划项目(2020GG0078);内蒙古自治区科技重大专项(2020ZD0009);内蒙古自治区高等学校科学技术研究项目(NJZY21460);内蒙古自治区自然科学基金项目(2019BS06003);内蒙古农业大学高层次人才科研启动项目(NDYB201857)。

摘　　要：针对传统叶片横切面宽度测量效率低、重现性差、劳动强度大,以及已有图像处理算法精度不高等问题,本研究提出一种基于变分水平集主动轮廓算法来提高叶片图像分割效果,从而更准确地测量细胞厚度。以狭叶锦鸡儿(Caragana Stenophylla Pojark)叶切片表皮细胞为研究对象,依据图像特点,首先采用基于几何均值和三维块匹配算法,实现图像盲去噪;然后采用变分水平集主动轮廓算法,分割预处理后图像提取表皮细胞;最后采用最小距离求交法确定出骨架曲线上局部拟合中心线的垂线与表皮细胞轮廓2交点,通过计算2交点间的距离,并除以对应比例关系,得出表皮细胞实际厚度。结果表明,①图像经去噪后,有效去除图像噪声,图像质量明显提高,相较传统Lazy Snapping算法和Graph Cut算法,本文提出算法虽耗时较长,但分割效果很明显,平均分割误差(R)、过分割误差(O_(R))和欠分割误差(U_(R))分别为5.670%、0.589%和5.900%,所用时间为30.228 s;②通过与交互测量值的比较得出,本文算法测量值与交互测量值间相对误差为3.27%,均方根误差为0.983,相对较小,结果较准确;③通过对同一幅图像30次重复测量后发现,相较ToupTek(高分辨率智能显微镜成像CCD相机)Toupview软件交互测量方法,本文算法用时更短,稳定性较高,误差为0.007。Aiming at the problems of low efficiency,high labor intensity,poor reproducibility and low accuracy of the existing image processing algorithms,in this study,an active contour algorithm based on variational level set was proposed to improve the segmentation effect of leaf image,so as to measure cell thickness more accurately.The epidermal cells of Caragana Stenophylla Pojark leaf slice were taken as the research object.Based on the image characteristics,the geometric mean and 3D block matching algorithm were firstly adopted to realize image blind denoising.Then the active contour algorithm based on variational level set was used to segment the pre-processed epidermal cell images and realize the extraction of epidermal cells.Finally,the fast marching skeleton extraction algorithm was used to determine the intersection of the vertical line of the local fitting center line on the skeleton curve and the epidermal cell contour.The actual thickness of epidermal cells was obtained by calculating the distance between the two intersections and dividing it by the corresponding proportional relationship.The results showed that:①After denoising,the image noise was effectively removed and the image quality was significantly improved.Compared with traditional Lazy Snapping algorithm and Graph Cut algorithm,the algorithm proposed in this paper took a long time,but the segmentation effect was obvious.Mean segmentation error(R),over-segmentation error(O_(R))and under-segmentation error(U_(R))value were 5.670%,0.589%and 5.900%,respectively,the average time was of 30.228 s.②Through the comparison with the interactive measured values,the relative error between the measured values and the interactive measured values of the algorithm in this paper was 3.27%,and the root mean square error was 0.983,which was relatively small and the results were relatively accurate.③After repeated measurement of the same image for 30 times,it was found that,compared with the interactive measurement method of ToupTek Toupview software,the algorith

关 键 词：厚度测量 变分水平集 主动轮廓 表皮细胞 狭叶锦鸡儿 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] S77[自动化与计算机技术—计算机科学与技术]