基于GLCM木材树种识别方法的研究
|
摘要
本文通过对国内外的数字图像处理技术在木材科学中研究以及对不同的识别技术的发展状况的分析,试验研究了基于木材灰度共生矩阵(GLCM)的数字化图像处理的方法,探讨了基于木材灰度特征进行木材树种识别的可行性。
不同树种的细胞类型、大小、排列、多少和壁腔比等都是不同的,这就导致了木材组织图像的灰度特征也不尽相同。在试验分析不同组织构造对木材灰度影响的基础上,针对灰度共生矩阵的特点,本文提取了木材横切面、弦切面上包括能量、熵、惯性矩、局部平稳性、相关等“纹理”特征(此“纹理”特征与木材纹理是两个不同的概念)。灰度共生矩阵参数由像素间距1到10进行讨论,结果表明像素间距越大,能量越小,熵越大,惯性矩越大,局部平稳性越小,相关越小。通过对不同像素间距的研究,找出了不同树种差异较大的像素间距。
本文分析了针阔叶树材在“纹理”特征值上的变异性,讨论了不同类型阔叶树材在“纹理”特征值上的变异性,比较研究了不同树种间特征值间的差异,并据此建立了66个树种的木材灰度特征值的数据库。
为了更好的研究木材“纹理”特征与木材组织构造的关系,本文对比研究了同种木材早晚材“纹理”特征值的差异性。通过二值化分析了木材的“胞壁率”、“壁腔比”,并分析了不同“胞壁率”、“壁腔比”的树种分布,通过比较表明阔叶树材的“胞壁率”较大。讨论了木材横切面上的“胞壁率”、“壁腔比”与木材“纹理”间参数的相关性。结果表明其相关性不大,说明了利用灰度来进行识别与传统的利用组织构造来识别有显著的不同。
通过对木材数字图像“纹理”灰度特征的探讨,本文列举了图像识别的几种方法,本文利用最小差值法进行判别树种。根据所识别的木材的特征与数据库中的特征进行比较,得出最接近树种,并给出木材的横切面、径切面、弦切面三切面图,识别模块同时给出了木材宏观特征和微观特征等基本信息。本文均利用Visual C++对程序进行实现。经过运行,结果表明,此方法可行。
In this paper, applying digital image processing technology about China and abroad in the wood science and showinig development of different identification skills ,proposing a digital image recognition method based on wood gray-level co-occurrence matrix(GLCM) , discussed the feasibility of using wood gray to identify the species.
Different wood species has different cell size, arrangement, more or less,ratio of cell wall are all different.By the influence of gray from different structure,aim at the feature of graly level co-occurrence matrix,This thesis extracted the wood texture features about cross section and tangential section, which including energy, entropy, moment of inertia, local stationarity,correlation(This texture feature and wood texure are two different concepts). Compared with the different texture feature. The parameters about gray-level co-occurrence matrix are discussed with 1 to 10 about the pixel distance.The results are when the bigger pixel distance,energy is lower,entropy is bigger,moment of inertia is bigger,local stationarity is bigger,correlation is lower. From the research of different pixel distance,find the pixel distance with a bigger difference among different species.
The paper anlysised the variability of softwood and hardwood in texture feature, discussed the variability of different kinds of hardwood in texture feature,anlysising the distinction between different species,and established a 66 species gray feature database from it.
In order to anlysis the influence of wood structure to texture parameter better,the paper analysised the distinction between earlywood and latewood by comparison.The paper analysis the ratio of cell wall,ratio of cell cavity ,ratio of cell wall to cavity based on binarization,and express the species distribution in different ratio of cell wall,ratio of cell wall to cavity.The ratio of cell wall of hardwood is relatively bigger by comparison.The paper discussed the correlation between cell wall ratio of cell cavity ,ratio of cell wall to cavity and texture parameters ,the result is the correlation is low,it shows the differences between using gray value to recognize and traditional using tissue structures to recognize.
This paper proposed several methods about image recognition,from the discussion of texture feature of wood digital image,this paper used minimum difference method to identify. we extracted the parameters about the identified species,compared with the fetures on database,we can get the most similar species,it also output 3 sections about cross section,radial section and tangential section.The recognition interface gives the macroscopic and microscopic characteristics and some other basical information. The paper all used Visual C++ to achive in programme.The result is the method is feasible.
不同树种的细胞类型、大小、排列、多少和壁腔比等都是不同的,这就导致了木材组织图像的灰度特征也不尽相同。在试验分析不同组织构造对木材灰度影响的基础上,针对灰度共生矩阵的特点,本文提取了木材横切面、弦切面上包括能量、熵、惯性矩、局部平稳性、相关等“纹理”特征(此“纹理”特征与木材纹理是两个不同的概念)。灰度共生矩阵参数由像素间距1到10进行讨论,结果表明像素间距越大,能量越小,熵越大,惯性矩越大,局部平稳性越小,相关越小。通过对不同像素间距的研究,找出了不同树种差异较大的像素间距。
本文分析了针阔叶树材在“纹理”特征值上的变异性,讨论了不同类型阔叶树材在“纹理”特征值上的变异性,比较研究了不同树种间特征值间的差异,并据此建立了66个树种的木材灰度特征值的数据库。
为了更好的研究木材“纹理”特征与木材组织构造的关系,本文对比研究了同种木材早晚材“纹理”特征值的差异性。通过二值化分析了木材的“胞壁率”、“壁腔比”,并分析了不同“胞壁率”、“壁腔比”的树种分布,通过比较表明阔叶树材的“胞壁率”较大。讨论了木材横切面上的“胞壁率”、“壁腔比”与木材“纹理”间参数的相关性。结果表明其相关性不大,说明了利用灰度来进行识别与传统的利用组织构造来识别有显著的不同。
通过对木材数字图像“纹理”灰度特征的探讨,本文列举了图像识别的几种方法,本文利用最小差值法进行判别树种。根据所识别的木材的特征与数据库中的特征进行比较,得出最接近树种,并给出木材的横切面、径切面、弦切面三切面图,识别模块同时给出了木材宏观特征和微观特征等基本信息。本文均利用Visual C++对程序进行实现。经过运行,结果表明,此方法可行。
In this paper, applying digital image processing technology about China and abroad in the wood science and showinig development of different identification skills ,proposing a digital image recognition method based on wood gray-level co-occurrence matrix(GLCM) , discussed the feasibility of using wood gray to identify the species.
Different wood species has different cell size, arrangement, more or less,ratio of cell wall are all different.By the influence of gray from different structure,aim at the feature of graly level co-occurrence matrix,This thesis extracted the wood texture features about cross section and tangential section, which including energy, entropy, moment of inertia, local stationarity,correlation(This texture feature and wood texure are two different concepts). Compared with the different texture feature. The parameters about gray-level co-occurrence matrix are discussed with 1 to 10 about the pixel distance.The results are when the bigger pixel distance,energy is lower,entropy is bigger,moment of inertia is bigger,local stationarity is bigger,correlation is lower. From the research of different pixel distance,find the pixel distance with a bigger difference among different species.
The paper anlysised the variability of softwood and hardwood in texture feature, discussed the variability of different kinds of hardwood in texture feature,anlysising the distinction between different species,and established a 66 species gray feature database from it.
In order to anlysis the influence of wood structure to texture parameter better,the paper analysised the distinction between earlywood and latewood by comparison.The paper analysis the ratio of cell wall,ratio of cell cavity ,ratio of cell wall to cavity based on binarization,and express the species distribution in different ratio of cell wall,ratio of cell wall to cavity.The ratio of cell wall of hardwood is relatively bigger by comparison.The paper discussed the correlation between cell wall ratio of cell cavity ,ratio of cell wall to cavity and texture parameters ,the result is the correlation is low,it shows the differences between using gray value to recognize and traditional using tissue structures to recognize.
This paper proposed several methods about image recognition,from the discussion of texture feature of wood digital image,this paper used minimum difference method to identify. we extracted the parameters about the identified species,compared with the fetures on database,we can get the most similar species,it also output 3 sections about cross section,radial section and tangential section.The recognition interface gives the macroscopic and microscopic characteristics and some other basical information. The paper all used Visual C++ to achive in programme.The result is the method is feasible.
引文
[1]陈兵棋,孙明.Visual C++实用图像处理[M].北京:清华大学出版社,2004.
[2]黄慧.基于木材组织构造的数字图像处理系统研究[D].南京.南京林业大学.2006.
[3]李盛阳.医学细胞图象分割与分析方法研究[D]硕士论文,2003.
[4]McMillin Charles W.Application of automatyi image analysis to wood science[J].Wood Sience.1982,14(3):97-105.
[5]ILIC J, HILLIS W E. Video image processor for wood automical quantification [J]. Holzforschung, 1983, 37 (1): 47-50.
[6]Peachey T.c&C.f.Osborne.The measurement of wood cell parameters using the Distance transform[J].Wood Fiber Science.1990,22(4):338-402.
[7]Fujita M.et al.Characteriztion of vessel distribution by FourierTransform Image Anlysis,Recent fast fourier transforms[J].Wood Anatomy,1995,(2):36-44.
[8]Diao X.M.et al.Analysis of cell arrangement in softwood by image processing using two-dimensional fast Fourier transforms[J].Wood Anatomy,1995,(1):211-215.
[9]Donaldson L.A.&M.J.F Lausberg.Comparison of conventional transmitted light and confocal microscopy for measuring wood cell dimensions by image analysis[J].IAWA Journal.1998,19(4):319-334.
[10]Mattias Moell &Gunilla Bargefors.An image anlaysis method of measure cross-sectional tracheid dimensions on softwood increment cores[J].Wood and Fiber Science,2001,33(2):201-212.
[11]Mattias K.Moell&Lloyd A.Donaldson.Comparison of segmentation methods for digital image analysis of confocal microscope images to measure tracheid cell dimensions.IAWA Journal,2001,22(3):267-288.
[12]王金满,邹常丰,郭军等.木材解剖构造图像处理的理论研究[J].东北林业大学学报,1994,22(2):52-57.
[13]王金满,刘一星,赵学增.木材构造计算机视觉分析方法[J].1993,21(2):94-99.
[14]王金满,张守娟,王宏威等.长白落叶松人工林木材构造计算机视觉分析的研究[J].东北林业大学学报.1995,23(3):51-56.
[15]张亚伟,王金满,王鹤滨等.长白落叶松林木材管胞径腔比的计算机视觉测量方法[J].林业科技.1996,21(2):31-32.
[16]王金满,曲艳杰,李坚.傅立叶变换图像处理方法在木材解剖特征研究上的应用[J].四川农业大学学报.1998,16(1):176-180.
[17]Wang Jinman,Qu Xiangjie,Wang Yongsheng.Analysis of Wood Anatomy Characteristics by Fast Fourier Transfer Image Analysis[J].Journal of Forestry Research,1997,8(4):243-245.
[18]孙丽萍,宋文龙,张冬妍.基于数字图像处理技术实现木材横纹压缩过程中构造学特征的研究[J].森林工程,2000,16(2):15-18.
[19]孙力平,曹军,李玉峰等.基于模式识别技术实现木材细胞图象边缘跟踪[J]林业机械与木工设备,1995,27(5):11-12.
[20]保昆雁,范晓宁.计算机视觉和图像处理技术与木材解剖特征量分析[C]中国林学会木材科学分会第九次学术研讨会论文集,2003,301-313.
[21]任洪娥,徐海涛.细胞特征参数计算机的提取理论[J].林业科学,2007,43(9): 68-73.
[22]王秀华.木材横切面构造特征计算机视觉分析与树种分类识别研究[D].哈尔滨:东北林业大学,2005.
[23]管雪梅.基于计算机视觉技术的木材细胞图像分割的研究[D].哈尔滨:东北林业大学,2006.
[24]李桂英.木材细胞图像特征提取及图像数据库的研究[D].哈尔滨:东北林业大学, 2005.
[25]杨家驹,程放.微机识别国产阔叶树材的研究.林业科学,1989,25(3):236-242.
[26]江泽慧,柯病凡等.中国主要商品微机识别的研究.安徽农学院学报,1990,16(2).
[27]程放,王艳君,陆熙娴.木材综合信息数据库查询系统的研究[J].木材工业,1995,9(4):16-19.
[28]徐杰,朱娅加.江西省木材数据库检索识别分类系统设计与实现[J].江西农业大学学报,1998,20(1):148-152.
[29]OHYAMAM,BABAK,ITOH I.Wood identification of Japanese Cyclobalanopsis species based on DNA Polymorphism of intergenic spacer between tm T and trn L 5 exon[J].Journal of Wood Science 2001,47:81-86
[30]马岩.木材横断面六棱规则细胞数字描述研究[J].生物数学学报,2002,17(1):64-68
[31]任洪娥.板材断面细胞实体检测图像的切片化处理研究[J].林业机械与木工设备,2006,34(3):26-27
[32]张宏林.精通Visual C++数字图像处理典型算法及实现[M].北京:人民邮电出版社,2008,7.
[33]Haralick R.,Shanmugam K.,Dinstein I.Textural Features for Image Classification,IEEE Transactions on Systems.Man and Cybernetics,1973,(6):610-621.
[34]刘明霞等.新的”纹理”图像特征提取方法[J].计算机应用,2009,12.
[35]苑丽红,孙爽滋,付丽针.灰度共生矩阵检索”纹理”图像的算法研究[J]计算机科学,2009,11.
[36]倾明,魏宗琴.图像检索中灰度共生矩阵的构造与实现[J]兰州石化职业技术学院学报,2009,3.
[37]徐小军,邵英,郭尚芬.基于灰度共生矩阵的火焰图像”纹理”特征分析[J]计算机技术与自动化,2007,12.
[38]尹思慈.木材学[M].北京:中国林业出版社,1996:25-28.
[39]Yan Ye.Text Image Compresssion Based on Pattern Matching[D].University of California,2002.
[40]Karadimitriou K,Fenstermacher M.Image compression in medical image databases using set redundancy[C].IEEE Proceedings of Data Compression Conference,1997,3:445.
[41]Karadunutriou K,Taler J M.The centroid method for compression sets of similar images[J].Pattern Recognition Letters,1998,19:585-593.
[42]Lee Jiann-Der,Wan Shu-Yen,Ma Cherng-Min,et al.Compression Sets of Similar Images Using Hybrid Compression Model[C].Proceedings 2002 IEEE International Conference on Multimedia and Expro,2002.
[43]胡小锋,赵辉.Visual C++/MATLAB图像处理与识别实用案例精选[M].北京:人民邮电出版社,2004,9:124-130.
[44]CONNERSRW,HARLOW C A.A theoretical comparison of texture algorithms[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1980,2(3):204-222.
[45]袁志发,周静芋.多元统计分析[M].北京:科学出版社,第一版,2003.
[46]王峰.基于语义数据模型的木材识别方法[J].南京林业大学学报2004,28(6):43-46
[47]王景寅,苏厚勤,腼立峰等.聚类算法在计算机木材识别系统中的应用[J].计算机工程,21-29.
[48]徐兆军,王厚立等.基于人工神经网络的原木CT图像缺陷识别[J]
[2]黄慧.基于木材组织构造的数字图像处理系统研究[D].南京.南京林业大学.2006.
[3]李盛阳.医学细胞图象分割与分析方法研究[D]硕士论文,2003.
[4]McMillin Charles W.Application of automatyi image analysis to wood science[J].Wood Sience.1982,14(3):97-105.
[5]ILIC J, HILLIS W E. Video image processor for wood automical quantification [J]. Holzforschung, 1983, 37 (1): 47-50.
[6]Peachey T.c&C.f.Osborne.The measurement of wood cell parameters using the Distance transform[J].Wood Fiber Science.1990,22(4):338-402.
[7]Fujita M.et al.Characteriztion of vessel distribution by FourierTransform Image Anlysis,Recent fast fourier transforms[J].Wood Anatomy,1995,(2):36-44.
[8]Diao X.M.et al.Analysis of cell arrangement in softwood by image processing using two-dimensional fast Fourier transforms[J].Wood Anatomy,1995,(1):211-215.
[9]Donaldson L.A.&M.J.F Lausberg.Comparison of conventional transmitted light and confocal microscopy for measuring wood cell dimensions by image analysis[J].IAWA Journal.1998,19(4):319-334.
[10]Mattias Moell &Gunilla Bargefors.An image anlaysis method of measure cross-sectional tracheid dimensions on softwood increment cores[J].Wood and Fiber Science,2001,33(2):201-212.
[11]Mattias K.Moell&Lloyd A.Donaldson.Comparison of segmentation methods for digital image analysis of confocal microscope images to measure tracheid cell dimensions.IAWA Journal,2001,22(3):267-288.
[12]王金满,邹常丰,郭军等.木材解剖构造图像处理的理论研究[J].东北林业大学学报,1994,22(2):52-57.
[13]王金满,刘一星,赵学增.木材构造计算机视觉分析方法[J].1993,21(2):94-99.
[14]王金满,张守娟,王宏威等.长白落叶松人工林木材构造计算机视觉分析的研究[J].东北林业大学学报.1995,23(3):51-56.
[15]张亚伟,王金满,王鹤滨等.长白落叶松林木材管胞径腔比的计算机视觉测量方法[J].林业科技.1996,21(2):31-32.
[16]王金满,曲艳杰,李坚.傅立叶变换图像处理方法在木材解剖特征研究上的应用[J].四川农业大学学报.1998,16(1):176-180.
[17]Wang Jinman,Qu Xiangjie,Wang Yongsheng.Analysis of Wood Anatomy Characteristics by Fast Fourier Transfer Image Analysis[J].Journal of Forestry Research,1997,8(4):243-245.
[18]孙丽萍,宋文龙,张冬妍.基于数字图像处理技术实现木材横纹压缩过程中构造学特征的研究[J].森林工程,2000,16(2):15-18.
[19]孙力平,曹军,李玉峰等.基于模式识别技术实现木材细胞图象边缘跟踪[J]林业机械与木工设备,1995,27(5):11-12.
[20]保昆雁,范晓宁.计算机视觉和图像处理技术与木材解剖特征量分析[C]中国林学会木材科学分会第九次学术研讨会论文集,2003,301-313.
[21]任洪娥,徐海涛.细胞特征参数计算机的提取理论[J].林业科学,2007,43(9): 68-73.
[22]王秀华.木材横切面构造特征计算机视觉分析与树种分类识别研究[D].哈尔滨:东北林业大学,2005.
[23]管雪梅.基于计算机视觉技术的木材细胞图像分割的研究[D].哈尔滨:东北林业大学,2006.
[24]李桂英.木材细胞图像特征提取及图像数据库的研究[D].哈尔滨:东北林业大学, 2005.
[25]杨家驹,程放.微机识别国产阔叶树材的研究.林业科学,1989,25(3):236-242.
[26]江泽慧,柯病凡等.中国主要商品微机识别的研究.安徽农学院学报,1990,16(2).
[27]程放,王艳君,陆熙娴.木材综合信息数据库查询系统的研究[J].木材工业,1995,9(4):16-19.
[28]徐杰,朱娅加.江西省木材数据库检索识别分类系统设计与实现[J].江西农业大学学报,1998,20(1):148-152.
[29]OHYAMAM,BABAK,ITOH I.Wood identification of Japanese Cyclobalanopsis species based on DNA Polymorphism of intergenic spacer between tm T and trn L 5 exon[J].Journal of Wood Science 2001,47:81-86
[30]马岩.木材横断面六棱规则细胞数字描述研究[J].生物数学学报,2002,17(1):64-68
[31]任洪娥.板材断面细胞实体检测图像的切片化处理研究[J].林业机械与木工设备,2006,34(3):26-27
[32]张宏林.精通Visual C++数字图像处理典型算法及实现[M].北京:人民邮电出版社,2008,7.
[33]Haralick R.,Shanmugam K.,Dinstein I.Textural Features for Image Classification,IEEE Transactions on Systems.Man and Cybernetics,1973,(6):610-621.
[34]刘明霞等.新的”纹理”图像特征提取方法[J].计算机应用,2009,12.
[35]苑丽红,孙爽滋,付丽针.灰度共生矩阵检索”纹理”图像的算法研究[J]计算机科学,2009,11.
[36]倾明,魏宗琴.图像检索中灰度共生矩阵的构造与实现[J]兰州石化职业技术学院学报,2009,3.
[37]徐小军,邵英,郭尚芬.基于灰度共生矩阵的火焰图像”纹理”特征分析[J]计算机技术与自动化,2007,12.
[38]尹思慈.木材学[M].北京:中国林业出版社,1996:25-28.
[39]Yan Ye.Text Image Compresssion Based on Pattern Matching[D].University of California,2002.
[40]Karadimitriou K,Fenstermacher M.Image compression in medical image databases using set redundancy[C].IEEE Proceedings of Data Compression Conference,1997,3:445.
[41]Karadunutriou K,Taler J M.The centroid method for compression sets of similar images[J].Pattern Recognition Letters,1998,19:585-593.
[42]Lee Jiann-Der,Wan Shu-Yen,Ma Cherng-Min,et al.Compression Sets of Similar Images Using Hybrid Compression Model[C].Proceedings 2002 IEEE International Conference on Multimedia and Expro,2002.
[43]胡小锋,赵辉.Visual C++/MATLAB图像处理与识别实用案例精选[M].北京:人民邮电出版社,2004,9:124-130.
[44]CONNERSRW,HARLOW C A.A theoretical comparison of texture algorithms[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,1980,2(3):204-222.
[45]袁志发,周静芋.多元统计分析[M].北京:科学出版社,第一版,2003.
[46]王峰.基于语义数据模型的木材识别方法[J].南京林业大学学报2004,28(6):43-46
[47]王景寅,苏厚勤,腼立峰等.聚类算法在计算机木材识别系统中的应用[J].计算机工程,21-29.
[48]徐兆军,王厚立等.基于人工神经网络的原木CT图像缺陷识别[J]