CHENG Yuzhi, ZHONG Lihui, HE Xin, et al. Segmentation of wood cross-section pores based on improved K-means clustering and watershed[J]. Journal of Zhejiang A&F University, 2022, 39(1): 173-179. DOI: 10.11833/j.issn.2095-0756.20210219
Citation: CHENG Yuzhi, ZHONG Lihui, HE Xin, et al. Segmentation of wood cross-section pores based on improved K-means clustering and watershed[J]. Journal of Zhejiang A&F University, 2022, 39(1): 173-179. DOI: 10.11833/j.issn.2095-0756.20210219

Segmentation of wood cross-section pores based on improved K-means clustering and watershed

DOI: 10.11833/j.issn.2095-0756.20210219
  • Received Date: 2021-03-11
  • Rev Recd Date: 2021-07-27
  • Available Online: 2022-02-14
  • Publish Date: 2022-02-14
  •   Objective  Pore is one of the important characteristics of wood identification. Aiming at the problems of poor robustness of pore segmentation due to random distribution and different sizes of pores and the impact of noise such as wood fiber, wood ray and axial parenchyma on the segmentation effect of pores, this study attempts to propose an improved K-means clustering and watershed algorithm for segmentation of wood cross-section pores.  Method  The improved K-means clustering was used to segment the pore area, which could distinguish the pore area from noise areas such as wood fiber, wood ray and axial parenchyma. Then, an improved watershed algorithm was used to segment the rough segmentation results, and the segmented pores were basically consistent with the actual pores.   Result  On average, 97.1% of the pores were segmented in each microscopic image of wood cross section. Compared with other algorithms, the improved algorithm had a significantly improved segmentation effect, high robustness and good segmentation performance in the process of pore segmentation with different sizes and random distribution.  Conclusion  This algorithm can effectively solve the problems of noise impact and randomness of initial clustering center in traditional K-means clustering algorithm in image segmentation, and lays a solid foundation for the feature extraction and quantitative analysis of hardwood pores. [Ch, 7 fig. 1 tab. 16 ref.]
  • [1] YANG Fan, YANG Bokai, LI Rongrong.  Surface defect detection technology of wood-based panel based on image segmentation and deep learning . Journal of Zhejiang A&F University, 2024, 41(1): 176-182. doi: 10.11833/j.issn.2095-0756.20230280
    [2] LUAN Jingran, FENG Guohong, ZHU Yujie.  Visible/near infrared spectrum wood identification based on SPA-GA-BP neural network . Journal of Zhejiang A&F University, 2022, 39(3): 671-678. doi: 10.11833/j.issn.2095-0756.20210377
    [3] LIU Yixiang, WANG Hangjun, XU Tieping.  A pest collection method based on global contrast . Journal of Zhejiang A&F University, 2021, 38(2): 369-376. doi: 10.11833/j.issn.2095-0756.20200318
    [4] LAN Jie, XIAO Zhongqi, LI Jimei, ZHANG Yutao.  Biomass allocation and allometric growth of Picea schrenkiana in Tianshan Mountains . Journal of Zhejiang A&F University, 2020, 37(3): 416-423. doi: 10.11833/j.issn.2095-0756.20190384
    [5] MING Hao, SU Xiyou.  Image recognition of poplar leaf diseases with feature segmentation and lesion enhancement . Journal of Zhejiang A&F University, 2020, 37(6): 1159-1166. doi: 10.11833/j.issn.2095-0756.20190752
    [6] ZHANG Guangqun, LI Yingjie, WANG Hangjun.  Classification of forestry images based on the BoW Model . Journal of Zhejiang A&F University, 2017, 34(5): 791-797. doi: 10.11833/j.issn.2095-0756.2017.05.004
    [7] BAI Xuebing, XU Jingtao, GUO Jingqiu, CHEN Kai.  Segmentation of wood surface knots and wormholes based on an improved LBF Model . Journal of Zhejiang A&F University, 2016, 33(2): 306-314. doi: 10.11833/j.issn.2095-0756.2016.02.017
    [8] YAO Fei, YE Kang, ZHOU Jianhua.  Automatic image classification and retrieval by analyzing plant leaf features . Journal of Zhejiang A&F University, 2015, 32(3): 426-433. doi: 10.11833/j.issn.2095-0756.2015.03.015
    [9] FENG Xue, WU Dasheng, LUO Wenjian.  A multi-depot emergency scheduling algorithm for forest fire prevention . Journal of Zhejiang A&F University, 2013, 30(2): 257-262. doi: 10.11833/j.issn.2095-0756.2013.02.016
    [10] JIANG Guangyu, XU Aijun, HUANG Xiaoyin, SHAO Jianlong.  Optimal path algorithm for forest fire prevention with constraints . Journal of Zhejiang A&F University, 2013, 30(1): 76-82. doi: 10.11833/j.issn.2095-0756.2013.01.011
    [11] HAO Hong, XU Chang-qing, ZHANG Xin-ping.  Aerial image information extraction based on non-negative matrix factorization . Journal of Zhejiang A&F University, 2012, 29(1): 72-77. doi: 10.11833/j.issn.2095-0756.2012.01.013
    [12] ZHANG Guang-qun, WU Wei-zhi, WANG Hang-jun.  A new wood microscopic image registration approach based on speeded up robust features (SURF) . Journal of Zhejiang A&F University, 2012, 29(4): 600-605. doi: 10.11833/j.issn.2095-0756.2012.04.018
    [13] XIA Guo-hua, LIU Chuan-he.  Sedum tricarpum:a new record of Sedum(Crasulaceae) from China . Journal of Zhejiang A&F University, 2011, 28(2): 254-255. doi: 10.11833/j.issn.2095-0756.2011.02.013
    [14] CHEN Jian-zhen, HE Chao, YUE Cai-rong.  Atmospheric correction of an advance land imager (ALI) image based on the FLAASH module . Journal of Zhejiang A&F University, 2011, 28(4): 590-596. doi: 10.11833/j.issn.2095-0756.2011.04.011
    [15] FANG Yi-ming, ZHENG Hong-ping, FENG Hai-lin.  Feature extraction and recognition of wood micrograph based on FFT and ICA . Journal of Zhejiang A&F University, 2010, 27(6): 826-830. doi: 10.11833/j.issn.2095-0756.2010.06.004
    [16] LIU Yi, SHE Guang-hui, LIU An-xing, ZHANG Guo-jiang.  Fuzzy cluster based on nature geographic regionalization of forest . Journal of Zhejiang A&F University, 2008, 25(4): 422-426.
    [17] WANG Bin, WANG Hui, YANG Jun-long, SUN Dong-yuan.  Regeneration characteristics for major tree species of a secondary Pinus tabulaeformis forest in the Ziwu Mountains . Journal of Zhejiang A&F University, 2007, 24(5): 559-563.
    [18] WANG Rui-ping, LIU Qiang, PENG Shao-lin, LIN Kai-hao, WEN Yan, XUE Ning.  Dynamics of microorganisms in litters of different tree species at Jianfengling . Journal of Zhejiang A&F University, 2006, 23(3): 255-258.
    [19] GUAN Yu.  Quartile division and regular-falsi-iteration algorithm of solving the equation . Journal of Zhejiang A&F University, 2001, 18(2): 184-187.
    [20] Zhang Junqi, Lin Shaoshen.  on Navel Orange introduction by Cluster Analysis of Ecoclimate in Zhejiang. . Journal of Zhejiang A&F University, 1996, 13(1): 41-47.
  • [1]
    HUANG Penggui, ZHAO Fan, LI Xiaoping, et al. Review of computer-based wood feature extraction and identification [J]. World For Res, 2020, 33(1): 44 − 48.
    [2]
    ESPINOSA L F, HERRERA R J, POLANCOTAPIA C. Segmentation of anatomical elements in wood microscopic images using artificial vision techniques [J]. Maderas Cienc Technol, 2015, 17: 106 − 108.
    [3]
    HUANG Peng, ZHENG Qi, LIANG Chao. Overview of image segmentation methods [J]. J Wuhan Univ Nat Sci Ed, 2020, 66(6): 519 − 531.
    [4]
    HAN Liuyang, TIAN Xingling, ZHOU Haibin, et al. The Influences of the anatomical structure and deterioration state of wood from a Qing Dynasty shipwreck on wood color after the consolidation treatment [J]. J Southwest For Univ Nat Sci, 2020, 40(1): 132 − 138.
    [5]
    YANG Jianfei, NING Liping, YANG Liao, et al. Variation in radial increment of Lindera megaphylla and its wood anatomical characteristics [J]. J Nanjing For Univ Nat Sci Ed, 2018, 42(1): 181 − 187.
    [6]
    Research Institute of Wood Industry, Chinese Academy of Forestry. A Wood Identification Method and System: CN201810830841.3[P]. 2019-01-03.
    [7]
    YU Haipeng, LIU Yixing, CUI Yongzhi, et al. Measurement of wood microstructural parameters on transverse section by binary morphology [J]. Mater Sci Technol, 2008, 16(1): 107 − 111.
    [8]
    JI Zhiwei, WANG Hangjun, HE Tao, et al. A morphological feature extraction method of wood pores based on an improved growing region algorithm [J]. J Beijing For Univ, 2011, 33(3): 64 − 69.
    [9]
    ZHANG Guangqun, WANG Hangjun. Multi-objective genetic based pore combination recognition [J]. Sci Silv Sin, 2012, 48(4): 87 − 92.
    [10]
    WANG Hangjun. The Study of Recognition Methods of Texture-based Wood Images[D]. Hefei: University of Science and Technology of China, 2013.
    [11]
    QI Hengnian, CHEN Fengnong, FANG Luming, et al. A method for wood pore image segmentation based on mathematical morphology [J]. J Beijing For Univ, 2008, 30(4): 12 − 16.
    [12]
    SHEN Pan, KUDO M. Segmentation of pores in wood microscopic images based on mathematical morphology with a variable structuring element [J]. Comput Electron Agric, 2010, 75(2): 250 − 260.
    [13]
    HE Tuo, GAO Ruiqing, JIAO Lichao, et al. Present status of global leading xylaria and suggestions for xylaria development in China [J]. Chin J Wood Sci Technol, 2020, 34(3): 40 − 43, 55.
    [14]
    LI Yugong. Improvement of K-means Clustering Algorithm and Its Application in Color Image Segmentation[D]. Hefei: Anhui University, 2018.
    [15]
    LU Yinju, DUAN Mingyi, LI Zuzhao, et al. An improved road crack image segmentation method [J]. Sci Technol Eng, 2020, 20(26): 10815 − 10821.
    [16]
    LI Ping. Morphological and Anatomical Studies on Fours Species of Acer L. [D]. Changchun: Jilin Agricultural University, 2011.
  • Created with Highcharts 5.0.7Amount of accessChart context menuAbstract Views, HTML Views, PDF Downloads StatisticsAbstract ViewsHTML ViewsPDF Downloads2024-052024-062024-072024-082024-092024-102024-112024-122025-012025-022025-032025-0400.250.50.7511.25Highcharts.com
    Created with Highcharts 5.0.7Chart context menuAccess Class DistributionFULLTEXT: 20.5 %FULLTEXT: 20.5 %META: 77.3 %META: 77.3 %PDF: 2.2 %PDF: 2.2 %FULLTEXTMETAPDFHighcharts.com
    Created with Highcharts 5.0.7Chart context menuAccess Area Distribution其他: 8.5 %其他: 8.5 %其他: 1.2 %其他: 1.2 %China: 0.4 %China: 0.4 %Ecuador: 0.3 %Ecuador: 0.3 %India: 0.1 %India: 0.1 %Rochester: 0.1 %Rochester: 0.1 %United States: 3.9 %United States: 3.9 %[]: 1.7 %[]: 1.7 %万隆: 0.4 %万隆: 0.4 %上海: 0.9 %上海: 0.9 %东莞: 0.2 %东莞: 0.2 %中卫: 0.1 %中卫: 0.1 %临汾: 0.1 %临汾: 0.1 %丽水: 0.1 %丽水: 0.1 %伊利诺伊州: 0.1 %伊利诺伊州: 0.1 %佛山: 0.1 %佛山: 0.1 %北京: 3.6 %北京: 3.6 %博阿努瓦: 0.1 %博阿努瓦: 0.1 %厦门: 0.1 %厦门: 0.1 %台州: 0.2 %台州: 0.2 %合肥: 0.1 %合肥: 0.1 %哥伦布: 0.1 %哥伦布: 0.1 %嘉兴: 0.1 %嘉兴: 0.1 %天津: 0.1 %天津: 0.1 %奥尔巴尼: 0.1 %奥尔巴尼: 0.1 %娄底: 0.1 %娄底: 0.1 %孟买: 0.3 %孟买: 0.3 %宣城: 0.1 %宣城: 0.1 %岳阳: 0.1 %岳阳: 0.1 %平顶山: 0.1 %平顶山: 0.1 %广州: 0.5 %广州: 0.5 %张家口: 1.2 %张家口: 1.2 %徐州: 0.1 %徐州: 0.1 %成都: 0.1 %成都: 0.1 %扬州: 0.8 %扬州: 0.8 %日照: 0.1 %日照: 0.1 %昆明: 1.8 %昆明: 1.8 %晋中: 0.2 %晋中: 0.2 %晋城: 0.1 %晋城: 0.1 %杭州: 1.7 %杭州: 1.7 %格兰茨维尔: 0.1 %格兰茨维尔: 0.1 %武汉: 0.3 %武汉: 0.3 %江门: 0.1 %江门: 0.1 %沈阳: 0.1 %沈阳: 0.1 %济南: 1.0 %济南: 1.0 %济宁: 0.1 %济宁: 0.1 %淮南: 0.3 %淮南: 0.3 %深圳: 0.4 %深圳: 0.4 %温州: 0.3 %温州: 0.3 %湖州: 0.4 %湖州: 0.4 %漯河: 1.0 %漯河: 1.0 %石家庄: 0.3 %石家庄: 0.3 %福州: 0.2 %福州: 0.2 %芒廷维尤: 19.7 %芒廷维尤: 19.7 %芝加哥: 0.2 %芝加哥: 0.2 %苏州: 0.1 %苏州: 0.1 %衡阳: 0.1 %衡阳: 0.1 %衢州: 0.1 %衢州: 0.1 %襄阳: 0.1 %襄阳: 0.1 %西宁: 33.3 %西宁: 33.3 %西雅图: 0.1 %西雅图: 0.1 %诺沃克: 7.5 %诺沃克: 7.5 %贝克斯菲尔德: 0.1 %贝克斯菲尔德: 0.1 %贵阳: 0.3 %贵阳: 0.3 %运城: 2.5 %运城: 2.5 %邯郸: 0.1 %邯郸: 0.1 %郑州: 0.3 %郑州: 0.3 %重庆: 0.1 %重庆: 0.1 %长沙: 0.5 %长沙: 0.5 %阳泉: 0.1 %阳泉: 0.1 %其他其他ChinaEcuadorIndiaRochesterUnited States[]万隆上海东莞中卫临汾丽水伊利诺伊州佛山北京博阿努瓦厦门台州合肥哥伦布嘉兴天津奥尔巴尼娄底孟买宣城岳阳平顶山广州张家口徐州成都扬州日照昆明晋中晋城杭州格兰茨维尔武汉江门沈阳济南济宁淮南深圳温州湖州漯河石家庄福州芒廷维尤芝加哥苏州衡阳衢州襄阳西宁西雅图诺沃克贝克斯菲尔德贵阳运城邯郸郑州重庆长沙阳泉Highcharts.com
  • Cited by

    Periodical cited type(1)

    1. 陆娇娇,段峰峰. 改进YOLOv5模型下农作物虫害图像识别算法仿真. 计算机仿真. 2023(11): 156-160 .

    Other cited types(4)

通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(7)  / Tables(1)

Article views(766) PDF downloads(23) Cited by(5)

Related
Proportional views

Segmentation of wood cross-section pores based on improved K-means clustering and watershed

doi: 10.11833/j.issn.2095-0756.20210219

Abstract:   Objective  Pore is one of the important characteristics of wood identification. Aiming at the problems of poor robustness of pore segmentation due to random distribution and different sizes of pores and the impact of noise such as wood fiber, wood ray and axial parenchyma on the segmentation effect of pores, this study attempts to propose an improved K-means clustering and watershed algorithm for segmentation of wood cross-section pores.  Method  The improved K-means clustering was used to segment the pore area, which could distinguish the pore area from noise areas such as wood fiber, wood ray and axial parenchyma. Then, an improved watershed algorithm was used to segment the rough segmentation results, and the segmented pores were basically consistent with the actual pores.   Result  On average, 97.1% of the pores were segmented in each microscopic image of wood cross section. Compared with other algorithms, the improved algorithm had a significantly improved segmentation effect, high robustness and good segmentation performance in the process of pore segmentation with different sizes and random distribution.  Conclusion  This algorithm can effectively solve the problems of noise impact and randomness of initial clustering center in traditional K-means clustering algorithm in image segmentation, and lays a solid foundation for the feature extraction and quantitative analysis of hardwood pores. [Ch, 7 fig. 1 tab. 16 ref.]

CHENG Yuzhi, ZHONG Lihui, HE Xin, et al. Segmentation of wood cross-section pores based on improved K-means clustering and watershed[J]. Journal of Zhejiang A&F University, 2022, 39(1): 173-179. DOI: 10.11833/j.issn.2095-0756.20210219
Citation: CHENG Yuzhi, ZHONG Lihui, HE Xin, et al. Segmentation of wood cross-section pores based on improved K-means clustering and watershed[J]. Journal of Zhejiang A&F University, 2022, 39(1): 173-179. DOI: 10.11833/j.issn.2095-0756.20210219
  • 阔叶材横截面微观图像由管孔、木射线、生长轮、木纤维和轴向薄壁组织构成,其中管孔是木材研究最重要的特征之一[1-3]。管孔不仅是分析考古木材样品的构造特征与保护状态的研究要素[3],也是研究某一地区若干年内木材生长量和环境气候变化相关性的依据[4]。管孔的分割是微观木材图像中提取管孔特征量和进行木材树种自动识别的关键步骤[56],因此,微观木材图像中的管孔分割有重要的研究意义。近些年来应用于管孔分割的算法主要有数学形态学、区域生长、目标遗传和水平集等[7-9]。张广群等[9]首先运用多目标遗传算法成功地分割出图像中的大部分管孔,但会出现复管孔或管孔团的漏分,同时部分轴向薄壁组织会连同其附近的管孔一起被分割出来。汪杭军[10]运用局部水平集算法,以局部灰度分布的均值作为统计信息,加强了管孔区域和背景区域的对比度,该算法虽然克服了漏分问题,但出现了噪声的误分割,一部分噪声为一些轮廓较大的木射线组织,另一部分为形状大小与管孔相似的木纤维组织。祁亨年等[11]运用改变结构元素和形态尺度的数学形态学算法对管孔进行分割,该算法克服了木射线和木纤维组织等噪声的影响,但出现了图像边缘细节丢失,如较小的管孔被当作噪声去除或是较大的管孔部分组织缺失。SHEN等[12]通过检测闭合区域的平均面积得出最适合的结构元素,并运用形态学算法对管孔进行分割,但还是会出现小孔漏分,大孔误分的情况。由于木材横截面微观图像中管孔数量众多、形态各异且随机分布,木纤维和轴向薄壁组织横截面也呈现一个个相对较小的孔洞,从而对管孔进行分割较为困难。综上所述,微观木材图像中管孔的分割依然是目前的研究热点和难点。本研究提出了一种改进K-means聚类与分水岭的木材横截面管孔分割算法,旨在解决管孔漏分、误分和噪声去除的问题并快速分割出管孔区域。

    • 日本森林数据库(https://db.ffpri.go.jp/WoodDB/JWDB/home.php)。该数据库成立于1928年,现有木材标本29 000余号,隶属270科2 050属8 500种;切片标本约90 000号[13]。本研究用到的木材树种为大红叶枫Acer amoenum(15帧图像)、尖齿槭A. argutum(5帧)、细柄槭A. capillipes(4帧)、茶条槭A. ginnala(1帧)、桑叶槭A. morifolium(3帧)、毛果槭A. nikoense(3帧)、日本槭A. nipponicum(5帧)、赛波德槭A. sicboldianum(24帧)、花楷槭A. ukurunduense(6帧)、红脉槭A. rufinerve(25帧)、薄叶槭A. tenuifolium(3帧)和褐枝猕猴桃Actinidia rufa(7帧)。

    • 传统的K-means聚类算法以距离作为相似性的评价指标,基本思想是按照距离将样本聚成不同的簇,两点距离越近,相似度就越大,以得到紧凑且独立的簇作为聚类目标[14-15]。传统的K-means聚类算法将样本映射到RGB颜色特征空间,构造新的映射关系,能够打破聚类只能是球状的限制,但聚类中心的选择会较大程度上影响分类效果,尤其是当聚类中心的选取全部位于管孔以外时,管孔区域会被整体归为噪声。此外,由于初始聚类中心的随机生成,可能会导致分类结果不一致或将木纤维或轴向薄壁组织被整体分割出来。基于此,针对以上问题进行算法改进,实现聚类中心的自适应选取。在RGB色彩空间下提取彩色木材微观图像的三通道分量。结合木材横截面原图,能够看出生长轮和木射线区域颜色较深,颜色值较小,所以其对应颜色值较小的峰值附近区域;管孔区域颜色较浅,颜色值较大,所以其对应颜色值较大的峰值附近区域;轴向薄壁组织和木纤维区域颜色较为平均,颜色值介于前两者之间,所以其对应两峰之间的波谷附近区域。分别取3个分量的均值组成第1个中心点,代表颜色直方图波谷附近区域,该点作为轴向薄壁组织和木纤维区域的簇类中心;遍历三通道分量,对于大于各自均值的所有颜色值取中值作为第2个中心点,由于该点对应颜色值较大,所以该点作为管孔的簇类中心;小于各自均值的所有颜色值取中值作为第3个中心点,由于该点对应颜色值较小,所以作为生长轮和木射线区域的簇类中心,分别代表颜色直方图两峰峰值附近区域。设通道分量为$ {M}_{j} $$ {M}_{j}\in (r,\;g,\;b) $Xn处的颜色值为In$ 0 < I $≤255。计算每个通道分量的均值$ \overline{M}_{j} $,均值两侧的中值为M1M2,Meidan表示取目标数组的中值函数。

      根据图像中各组织部分对应的颜色峰值或波谷得出这一组织部分的中心点,再以这3个点作为初始聚类中心点,运用K-means算法实现管孔的粗分割。

    • 虽然K-means聚类算法能够将木材微观图像中的管孔区域和背景区域区分开,但两者的颜色梯度差异较小,这可能会导致分水岭变换的过度分割。为了增加管孔区域和背景区域的颜色梯度差异,使分水岭变换在梯度运算判断极大值点时更加准确,更加有效地实现管孔的精分割,按照聚类的结果将木材微观图像进行3类颜色编码。第1类轴向薄壁组织细胞壁和木纤维区域进行红色编码,第2类管孔区域进行绿色编码,第3类生长轮和木射线区域进行蓝色编码。最后,通过形态学操作和分水岭变换实现管孔的精分割。

    • 为实现木材横截面微观图像的管孔分割,具体步骤如下(图1):①提取木材横截面微观图像的RGB分量并绘制颜色直方图;②根据木材微观图像中的3类目标区域,即轴向薄壁组织和木纤维、管孔、生长轮和木射线,确定聚类初始中心点;③根据像素点与聚类中心点之间的欧式距离求出新的簇类中心并判断中心点是否改变,若不一致则继续迭代,一致则进行下一步;④对划分出的3类区域进行颜色编码;⑤最后通过形态学操作去除面积较小的噪声,运用分水岭算法分割出管孔。

      Figure 1.  Improved K-means and watershed algorithm flow

    • 以大红叶枫横截面原图(997×1 418)为例,将本研究算法和现有算法(阈值分割、分水岭、K-means聚类与阈值分割)的分割效果进行对比分析,并以平均管孔有效分割率(F)作为算法分割性能的评价指标进行评价。

      式(4)~(5)中,Ti表示每张图像的管孔实际分割总数,Ni表示每张图像中的管孔总数,Pi表示每张图像的管孔有效分割率,重复次数n的取值为101,$ i\in n $F表示平均管孔有效分割率。

    • 随机选择3种木材微观横截面原图(图2)。在RGB色彩空间下提取彩色木材微观图像的三通道分量(图3)。可以看出3个分量均有2个峰和1个波谷组成,其中2个峰分别代表管孔区域和生长轮、木射线区域,波谷代表轴向薄壁组织和木纤维区域。

      Figure 2.  Original cross-section of wood

      Figure 3.  Color histogram of A. rufinerve

      按照颜色直方图得到初始聚类中心点。尖齿槭颜色值为(130.0, 144.0, 169.0),(20.0, 22.0, 49.0),(68.0, 78.4, 104.8);红脉槭颜色值为(84.6, 89.2, 110.4),(167.0, 184.0, 193.0),(32.0, 28.0, 56.0);赛波德槭颜色值为(181.0, 173.0, 207.0),(64.0, 40.0, 110.0),(113.0, 99.4, 154.9)。运用改进K-means聚类算法划分簇类(图4)。

      Figure 4.  Rough segmentation results of pores

      为了增加管孔区域和背景区域两者的颜色梯度差异,对聚类粗分割结果进行色彩编码(图5),第1类对轴向薄壁组织细胞壁和木射线区域进行红色编码,颜色值为(202, 12, 22);第2类对管孔区域进行绿色编码,颜色值为(53, 106, 195);第3类对生长轮以及木射线区域归进行蓝色编码,颜色值为(29, 209, 107)。

      Figure 5.  Color coding results

      经过灰度转换、开闭运算等图像形态学处理后,消除了绝大部分的木射线、木纤维和轴向薄壁组织等面积较小的噪声。此外运用分水岭变换对已分类的木材微观图像进行目标分割,即管孔分割(图6),将管孔快速准确地分割。从最终的分割结果来看,除了极个别的细小管孔未被检测到之外,其余管孔均被完整地分割,分割出的轮廓也与实际轮廓基本吻合。

      Figure 6.  Fine segmentation results of pores

    • 图7所示:阈值分割算法每张木材横截面微观图像平均有88.1%的管孔被准确有效地分割出来;分水岭算法每张木材横截面微观图像平均有91.4%的管孔被准确有效地分割出来,但存在噪声误分的问题;K-means聚类与阈值分割算法先运用K-means聚类算法将木材图像分为3类,再采用阈值分割和canny算子做管孔精分割,结果每张木材横截面微观图像平均有96.3%的管孔被准确有效地分割出来,但部分木射线和轴向薄壁组织也被分割出来,且存在轮廓不饱满的问题;本研究算法为改进K-means聚类与分水岭算法,结果每张木材横截面微观图像平均有97.1%的管孔被准确有效地分割出来,分割出的管孔与实际管孔基本吻合。

      Figure 7.  Segmentation effect diagram of each algorithm

      为验证本研究算法分割管孔的准确性,对12类树种的横截面微观图像在各算法下的分割结果进行了比较(表1),能够看出阈值分割不存在管孔误分,但因面积较小而漏分的管孔较多;分水岭算法不仅存在管孔漏分且会误分割出木射线、木纤维以及轴向薄壁组织;K-means聚类与阈值分割存在木射线和轴向薄壁组织的误分割;本研究算法基本不存在管孔的漏分和误分问题。

      算法管孔
      漏分
      木射线
      误分
      轴向薄壁
      组织误分
      木纤维
      误分
      阈值分割
      分水岭
      K-means聚类与阈值分割
      本研究算法

      Table 1.  Comparison of segmentation effects of various algorithms

    • 本研究改进的K-means聚类与分水岭的管孔分割算法,解决了初始中心点的随机性,并能有效地区分管孔区域与木纤维、木射线以及轴向薄壁组织等噪声区域,对管孔分割更为准确。本研究算法采用取图像颜色直方图对应的峰值确定初始聚类中心点和色彩编码的方式,充分将管孔与绝大部分噪声区域区分开,更适合于木材横截面微观图像分割任务。通过形态学处理和分水岭算法,对管孔进行精分割,将两者结合从而得到更为准确的分割结果。与其他分割算法相比,本研究算法分割效果显著提升,在大小不一且随机分布的管孔分割过程中鲁棒性高,有效避免了管孔漏分问题,具有良好的分割性能。

Reference (16)

Catalog

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return