圆钢端部图像识别的研究与处理
|
摘要
本文通过识别圆钢端部图像来实现对成捆圆钢的计数。首先，针对图像光照不均匀的特点，在Otsu自动域值选择方法的基础上，使用信息度量的方法对原始图像进行二值化处理。
其次，我们通过数学形态学的方法，针对图像噪声较强，背景较复杂的特点，使用了水线分割的方法，得到了目标图像的边缘图像。
然后通过灰度骨架变换方法和最大内接块的概念将图像中的颗粒噪声、目标表面及有噪背景分离开；移去变化背景及颗粒噪声，使图像得以增强；为了成功的识别和分离不需要的成分，用灰度形态开形成一个尺寸鉴定算法；在增强图像上应用全局阈值化从背景中获得目标。通过形态学基本方法的灵活运用，对目标图像的分割和背景噪声的消除取得了较好的效果。
最后，在Hough变换方法的基础上，根据系统要求响应速度快与内存需求较小的特点，使用了圆检测的RHT+算法。同时针对大部分的目标图形不是标准圆形结构的特点，采用模式特征向量及模式类分类决策的方法增强系统对圆形目标的识别能力，从而实现了圆钢计数的要求。
我们还提出了整个识别系统的设计方案，为以后的系统研究工作打下了基础。
In this dissertation, includes the contents of three parts mainly: basic process of binary image, object segmentation and background cut apart of image, circle object detection.
We are tying the question of counting of the round stock, Achieve the goal through recognizing the picture of round stock section. We use Otsu's automatic threshold selection method during the binarization in partitioned windows, a discriminating method based on the pyramid data structure of Lorentz information measure is put forward.
we propose a method based on mathematical morphology for segmenting images of round stock section, so we have received the edge picture of the goal picture.
In order to segment object and the background noise: First we separate the foreground speck noise and the noisy background in the image by means of the gray-scale skeleton transformation and the concept of maximal inscribed blocks. By removing the varying background and speck noise, the image is enhanced. To successfully recognize and separate the unwanted components, a size characterization algorithm is formulated based on the gray-scale morphological opening. Finally, a global thresholding can be applied to the enhanced image to obtain the object from the background.
There are a large amount of useless accumulations yielded by
random sampling when randomized Hough transform(RHT) is used to detect circles in complex images, so we proposes an improved RHT. It uses gradient direction information to determine whether the parameter based on the two sampled points should be accumulated or not. In comparison with the original RHT, the problem of useless accumulations is well solved and the method has higher speed, smaller storage. Direct against the object is not standard circle, we use the new algorithm combines two methods of cluster analysis and fuzzy recognition, recognizes step by step from coarse to fine, sets up classifier at every level. It solves the problem of uncertain objects classification successfully, implements recognition of irregular quasi-circular object accurately.
Finally, we put forward the plan of design of the whole recognition system, lay the foundation for the system's research work afterwards.
其次，我们通过数学形态学的方法，针对图像噪声较强，背景较复杂的特点，使用了水线分割的方法，得到了目标图像的边缘图像。
然后通过灰度骨架变换方法和最大内接块的概念将图像中的颗粒噪声、目标表面及有噪背景分离开；移去变化背景及颗粒噪声，使图像得以增强；为了成功的识别和分离不需要的成分，用灰度形态开形成一个尺寸鉴定算法；在增强图像上应用全局阈值化从背景中获得目标。通过形态学基本方法的灵活运用，对目标图像的分割和背景噪声的消除取得了较好的效果。
最后，在Hough变换方法的基础上，根据系统要求响应速度快与内存需求较小的特点，使用了圆检测的RHT+算法。同时针对大部分的目标图形不是标准圆形结构的特点，采用模式特征向量及模式类分类决策的方法增强系统对圆形目标的识别能力，从而实现了圆钢计数的要求。
我们还提出了整个识别系统的设计方案，为以后的系统研究工作打下了基础。
In this dissertation, includes the contents of three parts mainly: basic process of binary image, object segmentation and background cut apart of image, circle object detection.
We are tying the question of counting of the round stock, Achieve the goal through recognizing the picture of round stock section. We use Otsu's automatic threshold selection method during the binarization in partitioned windows, a discriminating method based on the pyramid data structure of Lorentz information measure is put forward.
we propose a method based on mathematical morphology for segmenting images of round stock section, so we have received the edge picture of the goal picture.
In order to segment object and the background noise: First we separate the foreground speck noise and the noisy background in the image by means of the gray-scale skeleton transformation and the concept of maximal inscribed blocks. By removing the varying background and speck noise, the image is enhanced. To successfully recognize and separate the unwanted components, a size characterization algorithm is formulated based on the gray-scale morphological opening. Finally, a global thresholding can be applied to the enhanced image to obtain the object from the background.
There are a large amount of useless accumulations yielded by
random sampling when randomized Hough transform(RHT) is used to detect circles in complex images, so we proposes an improved RHT. It uses gradient direction information to determine whether the parameter based on the two sampled points should be accumulated or not. In comparison with the original RHT, the problem of useless accumulations is well solved and the method has higher speed, smaller storage. Direct against the object is not standard circle, we use the new algorithm combines two methods of cluster analysis and fuzzy recognition, recognizes step by step from coarse to fine, sets up classifier at every level. It solves the problem of uncertain objects classification successfully, implements recognition of irregular quasi-circular object accurately.
Finally, we put forward the plan of design of the whole recognition system, lay the foundation for the system's research work afterwards.
引文
[1]陈岳林．CCD药片计数器的设计．计量与测试技术，2003，NO5：40-42
[2]王俊平，郝跃．IC真实缺陷的边界提取和缺陷尺寸与形状的表征．计算机学报，2000，VOL23(7)：673-378
[3]刘锡民，白丽华，张庆刚等．彩色切片细胞的识别与计数研究．山东师范大学学报(自然科学版)，2003，VOL18(2)：32-34
[4]戴青云，余英林．一种基于小波与形态学的车牌图像分割方法．中国图像图形学报，2000，VOL5(A)，N05：411-415
[5]杨海波，姚庆栋，荆仁杰．一种自然场景中的目标分割方法．中国图像图形学报，1998，VOL3，N08：646-650
[6]William K. Pratt. Digital Image Processing: PIKS Inside, Third Edition. John Wiley & Sons, Inc. 2001
[7]Rafael C. Gonzalez, Richard E. Woods. Digital Image Processing, Second Edition. Prentice Hall, Inc. 2002
[8]催屹．图象处理与分析—数学形态学方法及应用，北京：科学出版社，2000
[9]Littler J, Iuingworth J. Threshold selection based on a simple image statistics. Computer Graphics and Image Processing, 1985, 30(2): 125-147
[10]Abutaleb A S. Automatic thresholding of grey-level pictures using two-dimensional entropy. Computer Graphics and Image Processing, 1986, 47(1):22-32
[11]Otsu N. A Threshold selection method from gray-level histogram s[J]. IEEE Trans Systems, Man and Cybernetics, 1979,9(1): 62-66.
[12][美]张系国．图像信息系统设计原理．北京：科学出版社，1990：56
[13]黄庆明，张田文．基于Lorentz信息度量的分窗二值方法．航空学报，1995，VOL16，N05：614-618
[14]余农，吴常泳，李范鸣等．自动目标检测的形态学神经网络与模拟退火学习算法，中国科学(E辑)，2003，VOL33，N06：505-521
[15]张俊，刘立人，邵岚等．一种新的基于模糊推理的灰值形态学和它的光学实现，光子学报，1997，VOL26，N07：583-589
[16]余农，李予蜀，王润生．自动检测图像目标的形态滤波遗传算法，计算机学报，2001，VOL24，N04：337-346
[17]Gonzalez R C, Woods R E. Digital Image Processing. 3rd. ed., Addison-Wesley, 1992
[18]G.Matheron,Random. Sets and Integral in Geometry. Wiley, New York(1975)
[19]章毓晋．图像工程(上册)—图象处理和分析．北京：清华大学出版社，1999
[20]Lee C k, Wong S P. A mathematical morphological approach for segmenting heavily noise-corrupted images. Pattern recognition, 1996,29(8):1347-1357
[21]章毓晋．图像工程(下册)—图象理解与计算机视觉．北京：清华大学出版社，2000
[22]A. Rosenfeld, Segmentation:Pixel-based methods, Fundamentals in Computer Vision, O.D. Faugeras, ed.,pp. 225-238. Cambridge University Press, Cambridge, U.K.,1983
[23]R. Kohler, A segmentation system based on threshlding, Comput Graphics Image Process 1987, rOLl5:319-337
[24]K.S. Fu, R.C. Gonzalez and G.S.G. Lee, Robotics Control Sensing, Vision and Intellingence. McGraw-Hill, New York, 1987
[25]J.S. Wezka, A survey of threshold selection techniques, Comput Graphics Image Process 1978, VOL7:259-265
[26]J.C. Cheng, H.S. Don, Segmentation of bilevel images: a morphological approach, Pattern Recognition: Architecture, Algorithms and Applications, R. Plamondon and H.D. Cheng, eds, pp. 141-185. World Scientific Publishing Co., Singapore(1991)
[27]Xu L, Oja E.A New Curve Detection Method: Randomized Hough Transform(RHT).Pattern Recognition Letters, 1990,11(5):331-338
[28]Xu L, Oja E, Kultanen P. Randomized Hough rransform(RHT):Basic Mechanisms, Algorithms, and Computational Complexities. Computer Vision Graphics Image Process: Image Understanding, 1993,57(2):131
[29]韩海．从黑白图像中检测圆．武汉理工大学学报．VOL25，NO5：75-78
[30]许超．形态学准圆结构元素和骨架的研究．电子学报，VOL27，N08：78-81
[31]束志林，戚飞虎．一种新的随机Hough快速圆检测算法．计算机工程，2003，VOL29，N06(6)：87-88
[32]黄战华，李蒙，蔡怀宇等．一种视频图像规则圆目标的超分辨率位置测量方法．光学技术，VOL28，N02：125-127
[33]栾新，朱铁一．不规则类圆形目标图象识别新策略．中国图象图形学报，1999，VOL4(A)，NO3：202-206
[34]李伟青，彭群生．一种基于模式的圆的识别算法．软件学报，VOL10，NO2：
129-132
[35]尹令，黄辉先，汤红忠．车牌定位、分割算法研究及其在VC++中的实现．湘潭大学自然科学学报，2003，VOL25，NO2：102-105
[36]张琳，岑豫皖，杜培明．基于VC的棒材自动计数方法的研究与应用．工业控制计算机，2003，VOL16，NO3：32-34
[37]孙兆林，Matlab 6.x图像处理．北京：清华大学出版社，2002
[38]孙志永，蔡茂．谈软件开发工具的选择．中国计算机报，2002，Aug．8
[39]何斌，马天予，王运坚等．Visual C++数字图像处理．北京：人民邮电出版社，2001
[40]刘志敏 杨杰 施鹏飞，数学形态学的图像分割算法．计算机工程与科学，VOL20，NO4：21-27
[41]张培琨，李育林，刘家英等．基于小波边缘提取的灰度图像联合相关识别预处理．光子学报，VOL28，NO4：351-355
[42]赵唯，李端义．基于小波变换与形态学的一种边界重叠图像的拼接算法．微机发展，2001，NO6：3-6
[43]胡栋梁，叶芗芸，戚飞虎．数学形态滤波的细胞神经网络实现方法．上海交通大学学报，1997，VOL31，NO3：33-36
[44]李红松，侯朝桢．一种新的模糊边缘检测算法．计算机工程，2003，VOL29，NO9：1—2
[45]李艳军，左洪福，吴振锋等．灰色定权聚类在磨粒识别中的应用．数据采集与处理，2002，VOL17，NO1：108-112
[2]王俊平，郝跃．IC真实缺陷的边界提取和缺陷尺寸与形状的表征．计算机学报，2000，VOL23(7)：673-378
[3]刘锡民，白丽华，张庆刚等．彩色切片细胞的识别与计数研究．山东师范大学学报(自然科学版)，2003，VOL18(2)：32-34
[4]戴青云，余英林．一种基于小波与形态学的车牌图像分割方法．中国图像图形学报，2000，VOL5(A)，N05：411-415
[5]杨海波，姚庆栋，荆仁杰．一种自然场景中的目标分割方法．中国图像图形学报，1998，VOL3，N08：646-650
[6]William K. Pratt. Digital Image Processing: PIKS Inside, Third Edition. John Wiley & Sons, Inc. 2001
[7]Rafael C. Gonzalez, Richard E. Woods. Digital Image Processing, Second Edition. Prentice Hall, Inc. 2002
[8]催屹．图象处理与分析—数学形态学方法及应用，北京：科学出版社，2000
[9]Littler J, Iuingworth J. Threshold selection based on a simple image statistics. Computer Graphics and Image Processing, 1985, 30(2): 125-147
[10]Abutaleb A S. Automatic thresholding of grey-level pictures using two-dimensional entropy. Computer Graphics and Image Processing, 1986, 47(1):22-32
[11]Otsu N. A Threshold selection method from gray-level histogram s[J]. IEEE Trans Systems, Man and Cybernetics, 1979,9(1): 62-66.
[12][美]张系国．图像信息系统设计原理．北京：科学出版社，1990：56
[13]黄庆明，张田文．基于Lorentz信息度量的分窗二值方法．航空学报，1995，VOL16，N05：614-618
[14]余农，吴常泳，李范鸣等．自动目标检测的形态学神经网络与模拟退火学习算法，中国科学(E辑)，2003，VOL33，N06：505-521
[15]张俊，刘立人，邵岚等．一种新的基于模糊推理的灰值形态学和它的光学实现，光子学报，1997，VOL26，N07：583-589
[16]余农，李予蜀，王润生．自动检测图像目标的形态滤波遗传算法，计算机学报，2001，VOL24，N04：337-346
[17]Gonzalez R C, Woods R E. Digital Image Processing. 3rd. ed., Addison-Wesley, 1992
[18]G.Matheron,Random. Sets and Integral in Geometry. Wiley, New York(1975)
[19]章毓晋．图像工程(上册)—图象处理和分析．北京：清华大学出版社，1999
[20]Lee C k, Wong S P. A mathematical morphological approach for segmenting heavily noise-corrupted images. Pattern recognition, 1996,29(8):1347-1357
[21]章毓晋．图像工程(下册)—图象理解与计算机视觉．北京：清华大学出版社，2000
[22]A. Rosenfeld, Segmentation:Pixel-based methods, Fundamentals in Computer Vision, O.D. Faugeras, ed.,pp. 225-238. Cambridge University Press, Cambridge, U.K.,1983
[23]R. Kohler, A segmentation system based on threshlding, Comput Graphics Image Process 1987, rOLl5:319-337
[24]K.S. Fu, R.C. Gonzalez and G.S.G. Lee, Robotics Control Sensing, Vision and Intellingence. McGraw-Hill, New York, 1987
[25]J.S. Wezka, A survey of threshold selection techniques, Comput Graphics Image Process 1978, VOL7:259-265
[26]J.C. Cheng, H.S. Don, Segmentation of bilevel images: a morphological approach, Pattern Recognition: Architecture, Algorithms and Applications, R. Plamondon and H.D. Cheng, eds, pp. 141-185. World Scientific Publishing Co., Singapore(1991)
[27]Xu L, Oja E.A New Curve Detection Method: Randomized Hough Transform(RHT).Pattern Recognition Letters, 1990,11(5):331-338
[28]Xu L, Oja E, Kultanen P. Randomized Hough rransform(RHT):Basic Mechanisms, Algorithms, and Computational Complexities. Computer Vision Graphics Image Process: Image Understanding, 1993,57(2):131
[29]韩海．从黑白图像中检测圆．武汉理工大学学报．VOL25，NO5：75-78
[30]许超．形态学准圆结构元素和骨架的研究．电子学报，VOL27，N08：78-81
[31]束志林，戚飞虎．一种新的随机Hough快速圆检测算法．计算机工程，2003，VOL29，N06(6)：87-88
[32]黄战华，李蒙，蔡怀宇等．一种视频图像规则圆目标的超分辨率位置测量方法．光学技术，VOL28，N02：125-127
[33]栾新，朱铁一．不规则类圆形目标图象识别新策略．中国图象图形学报，1999，VOL4(A)，NO3：202-206
[34]李伟青，彭群生．一种基于模式的圆的识别算法．软件学报，VOL10，NO2：
129-132
[35]尹令，黄辉先，汤红忠．车牌定位、分割算法研究及其在VC++中的实现．湘潭大学自然科学学报，2003，VOL25，NO2：102-105
[36]张琳，岑豫皖，杜培明．基于VC的棒材自动计数方法的研究与应用．工业控制计算机，2003，VOL16，NO3：32-34
[37]孙兆林，Matlab 6.x图像处理．北京：清华大学出版社，2002
[38]孙志永，蔡茂．谈软件开发工具的选择．中国计算机报，2002，Aug．8
[39]何斌，马天予，王运坚等．Visual C++数字图像处理．北京：人民邮电出版社，2001
[40]刘志敏 杨杰 施鹏飞，数学形态学的图像分割算法．计算机工程与科学，VOL20，NO4：21-27
[41]张培琨，李育林，刘家英等．基于小波边缘提取的灰度图像联合相关识别预处理．光子学报，VOL28，NO4：351-355
[42]赵唯，李端义．基于小波变换与形态学的一种边界重叠图像的拼接算法．微机发展，2001，NO6：3-6
[43]胡栋梁，叶芗芸，戚飞虎．数学形态滤波的细胞神经网络实现方法．上海交通大学学报，1997，VOL31，NO3：33-36
[44]李红松，侯朝桢．一种新的模糊边缘检测算法．计算机工程，2003，VOL29，NO9：1—2
[45]李艳军，左洪福，吴振锋等．灰色定权聚类在磨粒识别中的应用．数据采集与处理，2002，VOL17，NO1：108-112