彩色套印偏差的检测方法研究
|
摘要
本文主要研究基于机器视觉的彩色套印偏差检测方法,据彩色套印偏差图像的边缘特征,提出了一种新颖的彩色套印偏差图像的检测方法——边缘共生条件概率矩阵。该算法可以用于控制印刷设备印刷的机械定位和离线分析彩色印刷品的印刷质量。本文详细研究和分析了各种经典的边缘检测算法并比较其各自的优势,选取了适合的边缘检测算法；介绍了本文提出的边缘共生条件概率矩阵算法和边缘共生条件概率矩阵极值点的提取算法；建立了一套彩色套印偏差图像检测算法流程。实验结果表明该算法对彩色套印偏差的检测是有效、精确的,检测精度可以精确到像素级。
引言部分阐述了本文的研究背景和意义,分析了现阶段国内外的研究现状,概述了本文的研究内容以及论文的结构安排。由于彩色套印是在纸张或织物同一位置的重叠印刷,而不是提前混合调配得到需要的颜色再印刷,这样就会存在不同的印刷颜色色版之间是否套准的问题。若色版套准超过质量的允许范围,则不仅会使印刷图案变得模糊,而且会引起色差,降低印刷品的质量。目前,无论是在线控制印刷设备的机械定位还是离线分析印刷质量,都是人工检测或者利用印刷品边缘的各种套准标记来进行检测的。人工检测不仅效率低而且检测精度不高；而利用套准标记不是直接针对于彩色印刷图案本身进行的,这样不仅在印刷前需要预留套准标记的位置而且完成印刷后需要裁剪该部分。本文在这样的背景下,研究了基于机器视觉的彩色套印偏差的检测方法,提出了直接针对于图像内容且不依赖模板图像的自动检测方法——边缘共生条件概率矩阵检测方法。
第二章研究了图像边缘检测的经典算法。图像边缘是图像中灰度变化比较剧烈的像素点的集合,它存在于目标与背景,目标与目标,区域与区域,基元与基元之间。边缘检测的经典检测算法有基于一阶微分的边缘检测方法,基于二阶微分的边缘检测方法,小波变换的边缘检测方法,基于数学形态学的边缘检测方法,基于分形理论的边缘检测方法等。经过分析对比,本文选取了Canny边缘检测作为彩色套印偏差图像的边缘检测方法,原因在于该方法具有高定位精度、低误判率、抑制虚假边缘的优点。
第三章提出了本文的核心算法——边缘共生条件概率矩阵。首先,讨论了重影图像的特点,彩色套印偏差图像是一种典型的重影图像。重影图像的重叠边缘可以认为是平行的,利用这个特点,本文提出了边缘共生条件概率矩阵的方法来统计彩色套印偏差图像的偏移参数。边缘共生条件概率矩阵利用不同角度、不同大小的扫描向量扫描彩色套印偏差图像,当扫描向量的起点和终点同时位于图像的边缘上时,则为该扫描向量对应的边缘共生条件概率矩阵元素值加一。若待检测图像存在彩色套印偏差,则在生成的边缘共生条件概率矩阵中会存在一个或者几个峰值,且该峰值点为彩色套印图像的偏差参数；若待检测图像无彩色套印偏差,则在生成的边缘共生条件概率矩阵中不存在明显的峰值点。
第四章讨论了基于边缘共生条件概率矩阵算法的彩色套印偏差的检测流程。首先讨论了彩色套印瑕疵的特点。在此基础上,介绍了彩色套印瑕疵检测的三个步骤：第一步,提出边缘图像,本文选取了Canny边缘检测的方法；第二步,在得到的边缘图像中,利用本文提出的边缘共生条件概率矩阵算法计算得到一个边缘公升条件概率矩阵；第三步,在得到的边缘共生条件概率矩阵中,去除其背景信息并计算提取峰值。
第五章讨论了实验设置,实验步骤及结果。本文的算法是用matlab语言实现的,选取了不同特点的彩色套印图像作为测试图像。实验结果表明本文提出的彩色套印偏差图像检测方法是精确有效的。
综上所述,本论文提出了一种无需模板图像,无需彩色套准标记,利用图像边缘信息的自动检测方法——边缘共生条件概率矩阵(ECM),用来检测彩色套印图像是否存在套印偏差及其偏差参数的估计。由于彩色套印偏差是由于某个或者某几个颜色印版的平移所造成的,所以本文采用对色版的平移信息的统计进行套印偏差的检测及其偏差参数的估计。经过实验,该方法对彩色套印偏差是精确、有效的,可以精确到像素级。
This paper carries out the research in the defects detecting of color printed overlapping based on computer vision technology. It presents a novel algorithm of Edge Co-occurrence conditional probability Matrix(ECM) according to the edge feature of color printed overlapping. It can be used to control the machinery positioning of printing equipment and analyze the quality of color printing offline. Firstly, this paper studies and analysis the edge detection algorithms in detail and selects one detection algorithm which is suitable for Edge Co-occurrence conditional probability Matrix(ECM). Then, it introduces the Edge Co-occurrence conditional probability Matrix(ECM) algorithm. At last, it introduces algorithm s about the extraction of image extreme point.
The preface presents the background and significance of this paper. It analysis the domestic and overseas research on color printed overlapping at present, summarizes the content and the organization of whole paper. Color overlay printing is printed by several different colors to the same location repeatedly in order to get the right color. In paper color printing, it mainly uses four colors, CMYK. Most part of colors in nature can be achieved by mixing different proportions of these four colors. In fabric dyeing, it uses much more colors to printing. The color overlay is the overlap printing on the same position in the paper or fabric rather than printing after get the pre-mixed colors. Because of this, it has the problem that if different print swatches have register problem. If the color register out of the mass range, it will induce the printing patterns become blurred and cause chromatic aberration, reduce the quality of products. By now, artificial detection or other detection methods which take advantage of registration marks at the edge of registration are used both in online printing or the offline analysis of the printing quality. But manual inspection is inefficient and imprecise. The methods that use registration marks are not detect the printed overlapping of color printing directly, it need to reserve the area for the marks and cutting after completion of the printing. To solve the problems above, this paper presents a solution for defects detecting of color printed overlapping based on computer vision technology, Edge Co-occurrence conditional probability Matrix(ECM). It is an automatic detection method for image content and does not depend on the template image.
The second chapter presents the image edge detection algorithms. The edge of image is the set of pixels whose gray transformation is dramatic in the image. It exists between the target and background, objectives and goals, region to region, primitive and the primitives. The classical edge detection algorithms includes the method based on the first derivative, the method based on second derivative, the edge detection method based on wavelet transform, the edge detection based on mathematical morphology, the edge detection based on Fractal Theory detection firm and so on. According to detailed analysis, this paper selects edge detection which uses Canny edge detection method to extract the color overprint deviation image. The main reason is this method can reduce the image weak edge information through two thresholds.
The third chapter proposes the core algorithm of this paper, Edge Co-occurrence conditional probability Matrix(ECM). Firstly, it discusses the characteristics of color overlay images. It is a typical ghost image. In the edge image of ghost image, its corresponding edge can be considered parallel. Using this feature, this paper proposes the Edge Co-occurrence conditional probability Matrix(ECM) to statistical parameters of color overlay image of migration. It uses different angles and different scan vector to scan the Image that to be detected. When the starting point of scan vector is located on the edge of the image, the corresponding element in the Edge Co-occurrence conditional probability Matrix(ECM) will plus one. If the image to be detected has color printed overlapping, its Edge Co-occurrence conditional probability Matrix(ECM) will have one or more peak points, and this or these points will be the deviation parameter of this color overlay images. If there is no color printed overlapping exist in the image, its Edge Co-occurrence conditional probability Matrix(ECM) has no obvious peak points.
The fourth chapter discussed the detection process of color printed overlapping based on the Edge Co-occurrence conditional probability Matrix(ECM). Firstly, it discusses the characteristics of color overlay defects. Then it proposed three steps for the detection of color printed overlapping:the first step is to extract the edge image, this paper selects the canny edge detection method; in the next step, calculated conditional probability of an edge matrix of symbiotic use the method that proposed in this paper; In step three, remove the background information and calculate the peak value of the matrix that get from step two.
The fifth chapter discusses the setup of experiment, the procedure of experiment, and the result. The algorithm proposed in this paper is implemented by matlab. It selects different color overlay images as test images. The experiment results show that the detection method proposed in this paper is accurate and efficient.
This paper presents a novel algorithm of Edge Co-occurrence conditional probability Matrix(ECM) for detecting color printed overlapping defects without the reference image and color mark using for registration. It can be used to detect the color printed overlapping defects and evaluate the shift parameter. Due to the color printed overlapping defects are usually caused by the shift of some certain printing plates, we use statistical shift information to detect the color printed overlapping defects and evaluate the shift parameter. Experiments in the paper demonstrate that the detection result can accurate to one pixel and the algorithm is effective and accuracy.
引言部分阐述了本文的研究背景和意义,分析了现阶段国内外的研究现状,概述了本文的研究内容以及论文的结构安排。由于彩色套印是在纸张或织物同一位置的重叠印刷,而不是提前混合调配得到需要的颜色再印刷,这样就会存在不同的印刷颜色色版之间是否套准的问题。若色版套准超过质量的允许范围,则不仅会使印刷图案变得模糊,而且会引起色差,降低印刷品的质量。目前,无论是在线控制印刷设备的机械定位还是离线分析印刷质量,都是人工检测或者利用印刷品边缘的各种套准标记来进行检测的。人工检测不仅效率低而且检测精度不高；而利用套准标记不是直接针对于彩色印刷图案本身进行的,这样不仅在印刷前需要预留套准标记的位置而且完成印刷后需要裁剪该部分。本文在这样的背景下,研究了基于机器视觉的彩色套印偏差的检测方法,提出了直接针对于图像内容且不依赖模板图像的自动检测方法——边缘共生条件概率矩阵检测方法。
第二章研究了图像边缘检测的经典算法。图像边缘是图像中灰度变化比较剧烈的像素点的集合,它存在于目标与背景,目标与目标,区域与区域,基元与基元之间。边缘检测的经典检测算法有基于一阶微分的边缘检测方法,基于二阶微分的边缘检测方法,小波变换的边缘检测方法,基于数学形态学的边缘检测方法,基于分形理论的边缘检测方法等。经过分析对比,本文选取了Canny边缘检测作为彩色套印偏差图像的边缘检测方法,原因在于该方法具有高定位精度、低误判率、抑制虚假边缘的优点。
第三章提出了本文的核心算法——边缘共生条件概率矩阵。首先,讨论了重影图像的特点,彩色套印偏差图像是一种典型的重影图像。重影图像的重叠边缘可以认为是平行的,利用这个特点,本文提出了边缘共生条件概率矩阵的方法来统计彩色套印偏差图像的偏移参数。边缘共生条件概率矩阵利用不同角度、不同大小的扫描向量扫描彩色套印偏差图像,当扫描向量的起点和终点同时位于图像的边缘上时,则为该扫描向量对应的边缘共生条件概率矩阵元素值加一。若待检测图像存在彩色套印偏差,则在生成的边缘共生条件概率矩阵中会存在一个或者几个峰值,且该峰值点为彩色套印图像的偏差参数；若待检测图像无彩色套印偏差,则在生成的边缘共生条件概率矩阵中不存在明显的峰值点。
第四章讨论了基于边缘共生条件概率矩阵算法的彩色套印偏差的检测流程。首先讨论了彩色套印瑕疵的特点。在此基础上,介绍了彩色套印瑕疵检测的三个步骤：第一步,提出边缘图像,本文选取了Canny边缘检测的方法；第二步,在得到的边缘图像中,利用本文提出的边缘共生条件概率矩阵算法计算得到一个边缘公升条件概率矩阵；第三步,在得到的边缘共生条件概率矩阵中,去除其背景信息并计算提取峰值。
第五章讨论了实验设置,实验步骤及结果。本文的算法是用matlab语言实现的,选取了不同特点的彩色套印图像作为测试图像。实验结果表明本文提出的彩色套印偏差图像检测方法是精确有效的。
综上所述,本论文提出了一种无需模板图像,无需彩色套准标记,利用图像边缘信息的自动检测方法——边缘共生条件概率矩阵(ECM),用来检测彩色套印图像是否存在套印偏差及其偏差参数的估计。由于彩色套印偏差是由于某个或者某几个颜色印版的平移所造成的,所以本文采用对色版的平移信息的统计进行套印偏差的检测及其偏差参数的估计。经过实验,该方法对彩色套印偏差是精确、有效的,可以精确到像素级。
This paper carries out the research in the defects detecting of color printed overlapping based on computer vision technology. It presents a novel algorithm of Edge Co-occurrence conditional probability Matrix(ECM) according to the edge feature of color printed overlapping. It can be used to control the machinery positioning of printing equipment and analyze the quality of color printing offline. Firstly, this paper studies and analysis the edge detection algorithms in detail and selects one detection algorithm which is suitable for Edge Co-occurrence conditional probability Matrix(ECM). Then, it introduces the Edge Co-occurrence conditional probability Matrix(ECM) algorithm. At last, it introduces algorithm s about the extraction of image extreme point.
The preface presents the background and significance of this paper. It analysis the domestic and overseas research on color printed overlapping at present, summarizes the content and the organization of whole paper. Color overlay printing is printed by several different colors to the same location repeatedly in order to get the right color. In paper color printing, it mainly uses four colors, CMYK. Most part of colors in nature can be achieved by mixing different proportions of these four colors. In fabric dyeing, it uses much more colors to printing. The color overlay is the overlap printing on the same position in the paper or fabric rather than printing after get the pre-mixed colors. Because of this, it has the problem that if different print swatches have register problem. If the color register out of the mass range, it will induce the printing patterns become blurred and cause chromatic aberration, reduce the quality of products. By now, artificial detection or other detection methods which take advantage of registration marks at the edge of registration are used both in online printing or the offline analysis of the printing quality. But manual inspection is inefficient and imprecise. The methods that use registration marks are not detect the printed overlapping of color printing directly, it need to reserve the area for the marks and cutting after completion of the printing. To solve the problems above, this paper presents a solution for defects detecting of color printed overlapping based on computer vision technology, Edge Co-occurrence conditional probability Matrix(ECM). It is an automatic detection method for image content and does not depend on the template image.
The second chapter presents the image edge detection algorithms. The edge of image is the set of pixels whose gray transformation is dramatic in the image. It exists between the target and background, objectives and goals, region to region, primitive and the primitives. The classical edge detection algorithms includes the method based on the first derivative, the method based on second derivative, the edge detection method based on wavelet transform, the edge detection based on mathematical morphology, the edge detection based on Fractal Theory detection firm and so on. According to detailed analysis, this paper selects edge detection which uses Canny edge detection method to extract the color overprint deviation image. The main reason is this method can reduce the image weak edge information through two thresholds.
The third chapter proposes the core algorithm of this paper, Edge Co-occurrence conditional probability Matrix(ECM). Firstly, it discusses the characteristics of color overlay images. It is a typical ghost image. In the edge image of ghost image, its corresponding edge can be considered parallel. Using this feature, this paper proposes the Edge Co-occurrence conditional probability Matrix(ECM) to statistical parameters of color overlay image of migration. It uses different angles and different scan vector to scan the Image that to be detected. When the starting point of scan vector is located on the edge of the image, the corresponding element in the Edge Co-occurrence conditional probability Matrix(ECM) will plus one. If the image to be detected has color printed overlapping, its Edge Co-occurrence conditional probability Matrix(ECM) will have one or more peak points, and this or these points will be the deviation parameter of this color overlay images. If there is no color printed overlapping exist in the image, its Edge Co-occurrence conditional probability Matrix(ECM) has no obvious peak points.
The fourth chapter discussed the detection process of color printed overlapping based on the Edge Co-occurrence conditional probability Matrix(ECM). Firstly, it discusses the characteristics of color overlay defects. Then it proposed three steps for the detection of color printed overlapping:the first step is to extract the edge image, this paper selects the canny edge detection method; in the next step, calculated conditional probability of an edge matrix of symbiotic use the method that proposed in this paper; In step three, remove the background information and calculate the peak value of the matrix that get from step two.
The fifth chapter discusses the setup of experiment, the procedure of experiment, and the result. The algorithm proposed in this paper is implemented by matlab. It selects different color overlay images as test images. The experiment results show that the detection method proposed in this paper is accurate and efficient.
This paper presents a novel algorithm of Edge Co-occurrence conditional probability Matrix(ECM) for detecting color printed overlapping defects without the reference image and color mark using for registration. It can be used to detect the color printed overlapping defects and evaluate the shift parameter. Due to the color printed overlapping defects are usually caused by the shift of some certain printing plates, we use statistical shift information to detect the color printed overlapping defects and evaluate the shift parameter. Experiments in the paper demonstrate that the detection result can accurate to one pixel and the algorithm is effective and accuracy.
引文
[1]金银河.展望21世纪我国的包装印刷[J].株洲工学院报,1999,13：16-18.
[2]陈昌杰等.塑料薄膜的印刷与复合[M].北京：化学工业出版社,1995：65-67.
[3]赵秀萍,许明飞.包装·设计·印刷[M].北京：印刷工业出版社,1995：89-91.
[4]Qkshun,xhffeng.套印[EB].http://baike.baidu.com/view/188922.html,2010,12.
[5]Srinivasan K,DastorP H,Radhakrishnaihan P, et al. FDAS:A Knowledge-based Frame Detection Work for Analysis of Defects in Woven Textile Structures[J]. The Journal of the Textile Institute,1992,83(3):431-447.
[6]吕志刚,曹跃进,刘重轩.新型圆网印花机对花精度检测系统的实现[J].江苏纺织,2003,7：50-52.
[7]万光逵,俞子荣,王琪.YS-1型微电脑凹版印刷自动套色装置[J].南京航空工业学院学报,1997,4：96
[8]Liulinghui,WangYue-zong.Study on color division model about registercalculation. Computer Engineering andApplications,2009,45(19): 210-212.
[9]刘京会,王跃宗.套准偏差计算分色归类方法研究[J].计算机工程与应用,2009,45(19)：210-212.
[10]YuLi_Jie, Li De-sheng, WangYue-zong.Application of image processing in detection of overprint deviation in colorprinting[J].Computer Engineering and Applications,2010,46(11):190-192.
[11]ZengXinxin, Li Desheng, Wang Yuezong.A new color image segment method for register detection[J].Micro-computer Imformation,2008.24.
[12]LiuHaoxue, YangWenjie, HUANG Min, et al. Detection and Control Algorithm of Multi- color Printing Registration Based on Computer Vision[C].Proceedings of The 2nd International Congress on Image and Signal Processing (CISP'09), Washington, USA:IEEE Computer Society,2009:2364-236.
[13]Gao Juan, DuanZhong-xin. Overprint error detection algorithm based on mathematical morphology. Journal of Computer Applications,2010.
[14]龙永红,吴敏.一种基于图像的套印参数测量方法[J].湖南大学学报(自然科学版).2006.4.
[15]徐建华.图像处理与分析[M].北京：科学出版社,1992.
[16]De Jong, J.N.M. and Castelli, V.R. et al.Method and apparatus for correction of color registration errors[P].US Patent 5,287,162.1994.
[17]Iwata, N., Sato, T. and Shinohara, T. et al. Image forming apparatus eliminating influence of fluctuation in speed of a conveying belt to correction of offset in color registration.US Patent 5,875,380.1999.
[18]A. Dockery. Automated fabric inspection:assessing the current state of the art[EB]. http://techexchange.com/thelibrary/FabricScanning.html, Jul.2001.
[19]Pratt W K. Digital Image Processing[M]. New York:Wiley,1991.
[20]章毓晋.图像处理和分析(上册)[M].北京：清华大学出版社,1999.
[21]Rafael C.Gonzalez, Richard E. Woods数字图像处理(第二版)[M].北京：电子工业出版社,2003.3：105.
[22]J. F. Canny, A computational approach to edge detection[J]. IEEE Trans. Pattern Anal. Machine Intell., vol. PAMI-8, pp.679-697,1986.
[23]郑南宁.计算机视觉与模式识别[M].北京：国防工业出版社,1998.
[24]M. J. Shensa,The discrete wavelet transform:Wedding theAtrous and Mallatalgorithm[J]. IEEETrans.SignalProcessing,vol.40, pp.2464-2482, 1992.
[25]StephaneMallat,Characterization of Signals from MultiscaleEdges[J]. IEEE IF,Vol.38, No.2, pp.617-643,1992.
[26]Witkin A. Scale Space Filtering[C]. ProcInt Joint ConfArtificial Intelligence Research,1983.
[27]Bergholm F. Edge Focusing[J]. IEEE Trans PAMI,1987,9:726-741.
[28]Mallat S. Zhong S. Characterization of Signals from MultiscaleEdges[J]. IEEE Trans PAMI,1992,14(7):710-732
[29]冯俊萍,赵转萍.基于数学形态学的图像边缘检测技术[J].航空计算技术,2004,34(3)：53-56.
[30]杨平先,孙兴波.一种改进多尺度形态边缘检测算法[J].光电工程,2005,32(11)：72-75.
[31]刘直芳.基于多尺度彩色形态矢量算子的边缘检测[J].中国图形图像学报,2002,7(9)：888-893.
[32]赵春晖,张乾,杨涛.基于数学形态滤波算子的医学图像边缘检测[J].信息技术,2002,16(11)：49-50.
[33]I. T. Young and L. J. Van Vliet. Recursive implementation of the Gaussian filters[J]. Signal Process,1995, vol.44:139-151.
[2]陈昌杰等.塑料薄膜的印刷与复合[M].北京：化学工业出版社,1995：65-67.
[3]赵秀萍,许明飞.包装·设计·印刷[M].北京：印刷工业出版社,1995：89-91.
[4]Qkshun,xhffeng.套印[EB].http://baike.baidu.com/view/188922.html,2010,12.
[5]Srinivasan K,DastorP H,Radhakrishnaihan P, et al. FDAS:A Knowledge-based Frame Detection Work for Analysis of Defects in Woven Textile Structures[J]. The Journal of the Textile Institute,1992,83(3):431-447.
[6]吕志刚,曹跃进,刘重轩.新型圆网印花机对花精度检测系统的实现[J].江苏纺织,2003,7：50-52.
[7]万光逵,俞子荣,王琪.YS-1型微电脑凹版印刷自动套色装置[J].南京航空工业学院学报,1997,4：96
[8]Liulinghui,WangYue-zong.Study on color division model about registercalculation. Computer Engineering andApplications,2009,45(19): 210-212.
[9]刘京会,王跃宗.套准偏差计算分色归类方法研究[J].计算机工程与应用,2009,45(19)：210-212.
[10]YuLi_Jie, Li De-sheng, WangYue-zong.Application of image processing in detection of overprint deviation in colorprinting[J].Computer Engineering and Applications,2010,46(11):190-192.
[11]ZengXinxin, Li Desheng, Wang Yuezong.A new color image segment method for register detection[J].Micro-computer Imformation,2008.24.
[12]LiuHaoxue, YangWenjie, HUANG Min, et al. Detection and Control Algorithm of Multi- color Printing Registration Based on Computer Vision[C].Proceedings of The 2nd International Congress on Image and Signal Processing (CISP'09), Washington, USA:IEEE Computer Society,2009:2364-236.
[13]Gao Juan, DuanZhong-xin. Overprint error detection algorithm based on mathematical morphology. Journal of Computer Applications,2010.
[14]龙永红,吴敏.一种基于图像的套印参数测量方法[J].湖南大学学报(自然科学版).2006.4.
[15]徐建华.图像处理与分析[M].北京：科学出版社,1992.
[16]De Jong, J.N.M. and Castelli, V.R. et al.Method and apparatus for correction of color registration errors[P].US Patent 5,287,162.1994.
[17]Iwata, N., Sato, T. and Shinohara, T. et al. Image forming apparatus eliminating influence of fluctuation in speed of a conveying belt to correction of offset in color registration.US Patent 5,875,380.1999.
[18]A. Dockery. Automated fabric inspection:assessing the current state of the art[EB]. http://techexchange.com/thelibrary/FabricScanning.html, Jul.2001.
[19]Pratt W K. Digital Image Processing[M]. New York:Wiley,1991.
[20]章毓晋.图像处理和分析(上册)[M].北京：清华大学出版社,1999.
[21]Rafael C.Gonzalez, Richard E. Woods数字图像处理(第二版)[M].北京：电子工业出版社,2003.3：105.
[22]J. F. Canny, A computational approach to edge detection[J]. IEEE Trans. Pattern Anal. Machine Intell., vol. PAMI-8, pp.679-697,1986.
[23]郑南宁.计算机视觉与模式识别[M].北京：国防工业出版社,1998.
[24]M. J. Shensa,The discrete wavelet transform:Wedding theAtrous and Mallatalgorithm[J]. IEEETrans.SignalProcessing,vol.40, pp.2464-2482, 1992.
[25]StephaneMallat,Characterization of Signals from MultiscaleEdges[J]. IEEE IF,Vol.38, No.2, pp.617-643,1992.
[26]Witkin A. Scale Space Filtering[C]. ProcInt Joint ConfArtificial Intelligence Research,1983.
[27]Bergholm F. Edge Focusing[J]. IEEE Trans PAMI,1987,9:726-741.
[28]Mallat S. Zhong S. Characterization of Signals from MultiscaleEdges[J]. IEEE Trans PAMI,1992,14(7):710-732
[29]冯俊萍,赵转萍.基于数学形态学的图像边缘检测技术[J].航空计算技术,2004,34(3)：53-56.
[30]杨平先,孙兴波.一种改进多尺度形态边缘检测算法[J].光电工程,2005,32(11)：72-75.
[31]刘直芳.基于多尺度彩色形态矢量算子的边缘检测[J].中国图形图像学报,2002,7(9)：888-893.
[32]赵春晖,张乾,杨涛.基于数学形态滤波算子的医学图像边缘检测[J].信息技术,2002,16(11)：49-50.
[33]I. T. Young and L. J. Van Vliet. Recursive implementation of the Gaussian filters[J]. Signal Process,1995, vol.44:139-151.