用户名: 密码: 验证码:
图像分割和边缘检测方法在昆虫图像中的应用
详细信息    本馆镜像全文|  推荐本文 |  |   获取CNKI官网全文
摘要
本文所研究的内容是昆虫识别的前期工作,为昆虫识别提供有利于特征提取的分割图像。由于昆虫的主要识别特征包括几何特征、颜色特征、纹理特征等,可以通过一系列的特征数据进行判断(如区域面积、边界周长、孔洞数、似圆度、偏心率等),这些数据间接反映了昆虫的许多自然特点,如身体的形状、颜色、体斑的多少、有无颚、触角及其形状如何等。因此需要首先对于昆虫图像进行分割操作,提取到便于进行特征数据计算的昆虫分割图像。主要进行三个步骤的操作并获得相应的分割图像:
     1、从复杂的彩色图像中分离昆虫对象。要研究的对象是昆虫,在提取昆虫的识别特征前应首先去除图像中除昆虫外的背景部分,以避免干扰。自然彩色图像的背景较为复杂,在去除时使用基于透明度的景物提取方法,本文中比较了Ruzon&Tomasi方法以及Poisson方法,并取得了实验结果,其结果图像的质量能够满足边缘检测和分割的要求。
     2、使用边缘检测和分割算法提取昆虫的边界图像和有较好斑特征或翅纹特征的分割图像。根据要提取的几何特征(区域面积、区域周长、偏心率、孔洞数、似圆度等)的要求,使用边缘检测方法获得昆虫的边界,以提取昆虫的整体形状,使用图像分割方法提取昆虫身体上的斑或翅纹,为计算孔洞数、似圆度等判断昆虫的斑特征和翅纹特征的数据提供了依据。在边缘检测算法的选择上,比较了几种常用的一阶微分算子的优劣,并在Canny算子的基础上提出了基于最大类间、类内距离比值的自动双阈值改进算法。在提取斑和翅纹的分割算法的选择上,对鳞翅目昆虫使用基于模糊集合的熵方法,为改善带有阴影的鞘翅目昆虫图像分割的效果,本文提出了基于局部阈值的最佳熵阈值分割方法。
     3、获得躯干和足、触角、颚的分离图像。一些昆虫在分类时只使用整体几何特征还不能够完全准确地进行分类,需要一些局部特征的辅助,如足、触角、颚等,并且对于鞘翅目昆虫来说,足、触角的摆放位置在一定程度上影响了整体几何特征数据的测量,因此需要将足、触角、颚与躯干进行分离。本文提出了基于数学形态学的自适应分割操作方法,采用了数学形态学中的开运算方法进行分离操作,在结构元素的选择上使用一种自适应的方式,根据昆虫实际情况确定结构元素,避免了躯干提取失真的现象,并减少了足、触角、颚的根部图像残留的现象。
     为在图像处理过程中去除噪声的干扰(尤其是由设备和传输引起的脉冲噪声和椒盐噪声的影响),首先对图像进行平滑操作,本文对常用的几种平滑滤波方法进行了分析,选择使用中值滤波方法,并使用边缘的方向信息对其进行了改进,以减少平滑过程中的细节损失。
This paper studies the earlier period work of insect recognition, which provides the segmentation image that is easier to extract the characteristics for the insect recognition. Insect's main recognition characteristics including the geometry characteristic, the color characteristic, the texture characteristic and so on, We may carry on the judgment through a series of characteristics data (for example region area, boundary perimeter, hole number, circularity, eccentricity and so on). These data have indirectly reflected insect's many natural characteristics, such as bodily shape, color, number of body spot, whether there is maxilla, antenna and its shape. Therefore we should first to carry on the segmentation operation regarding the insect image, which is advantageous for carries on the characteristic data computation. It mainly carries on three steps to obtain the corresponding segmentation image:
     1st, separates the insect object from the complex background color image. To avoid disturbance we should remove the insect's background in the image firstly. The background of nature color image is complex. We use the matting method based on the transparency scenery in this paper, compared Ruzon & Tomasi method as well as the Poisson method, and have obtained the experimental result. The result could satisfy the edge detection and the segmentation request.
     2nd, using the edge detection and the segmentation algorithm obtain the insect's boundary image and the image which has better spot characteristic or the wing grain characteristic segmented image. According to the geometry characteristic which must extract (region area, region perimeter, eccentricity, hole number, circularity and so on), we could obtain the insect's boundary that express insect's overall shape with edge detection method and get the spot or the wing grain of the insect body with image segmentation method, which is the basis for calculating the hole number, circularity etc spot characteristic and the wing grain characteristic data. In this paper we've compared several predicate of first order operators (Roberts, Sobel, Prewitt, Canny), and proposed an improved algorithm which is based on the automatic double threshold value improvement algorithm based on the maximum inter-cluster distance and intra-cluster variance ratio. To withdrawing the spot and wing grain, a segmentation algorithm that based on the fuzzy set entropy method is used to Lepidoptera. And there is a best entropy threshold value segmentation method based on the partial threshold value that is proposed in this paper.
     3rd, obtain the segmented image of the trunk and the other part of insect. Some insects can not be classified completely accurately by the overall geometry characteristic only, so some partial characteristics such as foot, antenna and so on is needed. But to some insects, the position of the foot and antenna may affect the survey of the overall geometry characteristic data in a certain degree. So a separating image is needed. In this paper an open operational method of mathematics morphology is used to carry on the procession. At the same time, an adaptive method is proposed to choose the structural element size. This improvement could avoid distort phenomenon of the trunk abruption, and reduced remaining phenomenon of the foot, the antenna or maxilla's root.
     To eliminate the disturbance of the noise (particularly the pulse noise and salt-and-pepper noise influence which causes by equipment and transmission) in image processing, a smooth operation is needed. In this paper we have analyzed several smooth operation methods and improve median filter method by a directional way to reduce the detail loss of smoothing process.
引文
1.沈佐锐,于新文.温室白粉虱自动计数技术研究初报[J],生态学报,2001,21(1):94-99.
    2.于新文,沈佐锐.昆虫数字图像的分割技术研究[J].农业工程学报,2001,17(3):137-141.
    3.杨宏伟,张云.计算机视觉技术在昆虫识别中的应用进展[J].生物信息学,2005,3(3):133-136.
    4.付承彬,詹希美等.数字图像处理技术在并殖吸虫自动分类中的应用研究[J].中国人兽共患病杂志,2002,18(1):39-43.
    5.黄小燕,郭勇,赵太飞.数学形态学的储粮害虫彩色数字图像分割[J].计算机测量与控制,2003,11(6):467-469.
    6.陈兵旗,孙明.Visual C++实用图像处理专业教程[M].北京:清华大学出版社,2004,54-113.
    7.周红,王宏坡.基于Visual C++.NET的昆虫图像自动识别系统的研究[J].天津农学院学报,2005,12(2):39-41,53.
    8.陈健能,张国凤.计算机视觉技术在农业中的应用及展望[J].甘肃农业大学学报,2003,38(2):248-253.
    9.张学庆,刘燕.数学形态学在储粮昆虫图像识别中的应用[J].武汉工业学院学报,2004,23(1):94-96.
    10.鲁瑞华,张为群.模糊加权中值滤波器[J].计算机科学,2006,33(6):186-187,191.
    11.郭晓新,卢奕南,许志闻,王云霄,庞云阶.自适应定向加权中值滤波[J].吉林大学学报,2005,43(4):494-498.
    12.王志刚,王伟,许晓鸣.一种改进型自适应加权模糊均值滤波算法[J].高技术通讯,2003,13(2):19-24.
    13. T.Porter, T.Duff. Compositing digital images[J]. In SIGGRAPH 84, July 1984, 10(8): 253-259.
    14. Mark A.Ruzon, Carlo Tomasi. Alpha Estimation in Natural Images[J]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2000, 2000, 1(1): 18-25.
    15. P.Hillman, J.Hannah, D.Renshaw. Alpha channel estimation in high resolution images and image sequences[J]. In CVPR 2001, 2001, 20(5): 1063-1068.
    16. Jian Sun, Jiaya Jia, Chi-Keung Tang, Heung-Yeung Shum. Poisson Matting[J]. ACM Transaction on Graphics,2004, 23(3): 315-321.
    17. Michael Orchard, Charles Bouman. Color Quantization of Images[J]. IEEE Trans. On Sig. Proc., 1991,39(12):2677-2690
    18. 吴健辉,张国云,袁帅.一种改进的基于α值的景象提取算法研究[J].现代电子技术,2005,28(22): 57-59.
    19. Yung-Yu Chuang, Brian Curless, David H. Salesin, Richard Szeliski. A Bayesian Approach to Digital Matting[J]. In Proceedings of IEEE Computer Vision and Pattern Recognition,2001,2(7):264-271.
    20.李弼程,彭天强,彭波等.智能图像处理技术[M].北京:电子工业出版社,2004,149-156.
    21.郑翔,黄艺云.经典边缘检测模板的快速算法[J].信号处理,1995,11(4):317-320.
    22. P. Meer, B. Georgescu. Edge detection with embedded confidence[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2001,23 (12):1351-1365.
    23.周龙,卢亚玲.储粮昆虫图像边缘检测算法研究[J].计算机应用研究,2005,15(5):140-141,216.
    24.万力,易昂,傅明.一种基于Canny算法的边缘提取改善方法[J].计算技术与自动化,2003,22(1):24-26.
    25. PL. Worthington. Enhanced Canny edge detection using curvature consisitency[J]. Proceedings of 16th International Conference on Pattern Recongnition, 2002, 1 : 596- 599.
    26.韦海萍,赵保军,唐林波,何佩琨.Canny算法的改进及其硬件的实现[J].光学技术,2006,32(2):263-266.
    27.梁光明,孙即祥,马琦,姚春光.Otsu算法在Canny算子中的应用[J].国防科技大学学报,2003,25(5):36-39.
    28.祝轩,耿国华,周明全等.双阈值法在图像边缘检测中的应用[J].微机发展,2005,15(9):133-132.
    29.王植,贺赛先.一种基于Canny理论的自适应边缘检测方法[J].中国图像图形学报,2004,9(8):957-962.
    30.吴薇,刘军,李旭霞.图像分割中的阈值选取方法[J].西安工业学院学报,2002,22(4):17-22.
    31.罗涛华.基于自适应阈值的储粮害虫图像分割算法[J].武汉工业学院学报,2006, 25(1):5-8.
    32.蒋刚毅,郑义,郁梅,陈燕华.关于彩色图像处理的形态学算子研究[J].中国图像图形学报,2000,5(4).
    33. Liang Tang, Wei-Xin Xie, Jian-Jun Huang, Automatic Multilevel Image Segmentation Based on Fuzzy Reasoning[J]. Image anal Stereol,2004,23(1):23-31.
    34. Ping-Sung Liao, Tse-Sheng Chen, Pau-Choo Chung. A Fast Algorithm for Multilevel Thresholding[J]. Journal of Information Science and Engineering,2001,17(5):713-727.
    35.于新文,沈佐锐,高灵旺,李志红.昆虫图像几何形状特征的提取技术研究[J].中国农业大学学报,2003,8(3):47-50.
    36. Ming Tang, Song-De MA. Enhancement and Segmentation of Images in Computer Vision[J]. Journal of the Graduate School of the Chinese Academy of Sciences,2004,21(1): 140-143.
    37.张爱华,余胜生,周敬利.一种基于边缘检测的局部阈值分割方法[J].小型微型计算机系统,2003,24(4):661-663.
    38.俞庆英,王虎,吴建国.对图像分割技术的研究与实例图像测试[J].微机发展,2005,15(10):150-151.
    39.王颖,张广军,陈大志.昆虫翅膀图像特征的亚像素级提取方法[J].北京航空航天大学学报,2005,31(8):839-842.
    40.徐昉,邱道尹,沈宪章.粮仓害虫的特征提取与分类的研究[J].郑州工业大学学报,2000,21(4):62-65.
    41.张红梅,范艳峰,田耕.基于数字图像处理技术的储粮昆虫分类识别研究[J].河南工业大学学报,2005,26(1):19-22.
    42.崔屹.图像处理与分析—数学形态学方法及应用[M].北京:科学出版社,2000,15-28.
    43.欧珊瑚,王倩丽,朱哲瑜.Visual C++.NET数字图像处理技术与应用[M].北京:清华大学出版社,2004,244-472.
    44. Lei Wang, Ting-Zhi Shen. Two-Dimensional Entropy Method Based on Genetic Algorithm[J]. Journal of Beijing Institute of Technology, 2002,11 (2): 184-199.
    45. Xiaoyan Dai, J. Maeda. Fuzzy-based segmentation of color natural images[J]. 6th International Conference on Signal Processing, 2002, 2: 969-972.
    46.雷艳敏,黄秋元.基于数学形态学的图像边缘检测[J].武汉理工大学学报,2005,27(5):25-27.
    47.刘燕.储粮昆虫图像模式识别研究[D].四川成都:成都理工大学,2002:1-50.

© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号

地址:北京市海淀区学院路29号 邮编:100083

电话:办公室:(+86 10)66554848;文献借阅、咨询服务、科技查新:66554700