基于图像处理的目标特征识别算法研究
|
摘要
随着硬件设备和图像处理技术的不断提高发展,目标特征识别技术也在快速发展,其基本原理是利用图像中的目标特征信息,通过各种试图像处理技术来改善图像质量,提取特征,最后根据大量训练样本所确定的模板图像,代替人对被检测图像进行识别判决。目标特征识别虽然在理论方法研究上已取得不小的进展,但它本身是一个比较复杂、繁琐、要求精细的研究领域,存在精度与速度等不少困难,因此目标特征识别是一个有待进一步探索的新领域。
本文依据企业的印刷品质量检测的要求,描述了基于图像处理的目标特征识别问题的一个完整的实现和改进,改进了Canny算法进行边缘检测,同时应用了改进的可靠的模板配准算法,配以局部的动态阈值分割处理来获得识别结果,获得了准确的质量检测结果。
首先研究了图像获取的硬件系统,这是进行基于图像处理的目标特征识别研究的基础。选取设备的主要因素是设备的性能效果,同时也要仔细考虑环境因素,以及各设备之间的匹配融合,还有就是要协调考虑软件的设计,这样才能设计出最合适的系统。
其次,每幅图像都包含某种程度的噪声,综合考虑速度与效果,选用高斯滤波器来对图像进行平滑处理,对图像中的噪声进行抑制；而且应用图像分割处理技术,将图像分割为二值类型的图像,将来就可以忽略背景而直接对目标物体进行处理。在图像分割中,使用细致的、基于局部处理的动态阂值分割处理,可以得到准确的、稳定的处理结果。这样对原始图像进行了预处理,从而获得了质量较好的图像,尽力保证最终目标特征识别结果的准确性。
然后,因为如果对边缘进行处理比对整个目标物体处理的话,边缘信息的准确率远远高于整个目标物体的信息,所以需要对目标物体进行边缘提取。本文研究了Canny边缘检测算法,并在此基础上针对传统Canny算法在梯度幅值计算上对噪声过于敏感、容易检测出伪边缘的缺陷,应用了一种新的计算梯度幅值的算法,大大提高了准确度。然后针对传统Canny算法需要人为指定双阂值的缺陷,应用了一种根据非最大抑制处理的结果图像、来自动生成双阈值的方法,更全面地利用了梯度图像中的信息特征,使算法具有了更好的自适应能力,提高了自动化程度。
还有,当基于模板来检测物体时,要使图中物体的位姿与参考图像中物体的位姿相同。要进行模板配准,就是要在目标物体图像中找到模板,得到模板在图像中最合适的位姿,然后依据位姿变换关系,进行图像变换,将模板与图像中的目标物体对齐。先研究了传统的基于灰度值的模板配准算法的适用情况和使用效果,最后应用了改进的可靠的基于边缘形状的模板配准算法,配以局部的动态阂值分割处理来获得识别结果,算法准确度大大提高,也增加了对于各种质量的图像的适用范围。然后,在图像变换中,使用双线性插值变换方法,可以使得到的图像的变换结果具有平滑的外观边缘。
最后,模板与目标物体图像对齐后,质量检测算法使用的是动态阈值分割处理,设计了含有容许偏差的偏差模板,然后将与图像中的目标物体与其进行比较,以实现质量检测,结果十分可靠。
Target feature recognition is currently the focus area and hot spot in the computer vision and artificial intelligence research, which has a very wide range of applications in the military field, intelligent transportation, robot navigation, intelligent monitoring, traffic management, industrial inspection, and fingerprint identification, face recognition, iris recognition and other fields. The basic principle of the target feature recognition is that by using the target image's feature information, improving image quality with various image processing techniques, and then extracting features, finally to identify and the recognition images instead of people, based on templates images from a large number of training samples. Meanwhile, image processing technology has been paid a widespread attention to in more application fields, and has become a new potential subject with bright prospects.
With the fast development of computer technology and notice science, digital image
Processing is now developing in a deeper level. People begin the study to analyze and interpret images by using computer systems, realizing the outside world recognition with the vision systems similar to human beings'. Though having got some obvious progress in theory and Methodology, target feature recognition is still a new field needing more exportation for its complexity and high requirements,
This paper mainly studies the target feature recognition method based on the image processing. It compares the various template matching algorithms, and uses the image edge gradient vector characteristics normalized and improved. In this paper, a reliable template matching algorithm based on edge shapes is proposed, and the local dynamic threshold segmentation processing is used to obtain recognition with accurate results and high stability.
Firstly, this paper chooses gauss filter to smooth the image, rejecting the noise of the image. In the image transformation, the use of bilinear interpolation transform method can obtain the transformation results with a smooth appearance edge. In the image segmentation, the use of meticulous processing based on local processing dynamic threshold segmentation can get the exact and stable processing results. The original image preprocessed with good quality can ensure the accuracy of the ultimate target feature recognition results.
Then the Canny edge detection algorithm is studied in this paper. Because the traditional Canny algorithm for calculating the gradient amplitude is too sensitive to the noise and easily causes the false-edge detect, a new algorithm for computing gradient magnitude is proposed. With the disadvantages that the traditional Canny algorithm requires man-made dual threshold, a approach is proposed. The approach can automatically generate two-threshold according to the non-maximal inhibition image, and can make a more comprehensive use of information features of the gradient image. The proposed approach make the algorithm have better adaptive capacity, and improve the degree of automation.
At last, with the Contrastive study on the usage and effects of template matching algorithms based on the gray value, edge pixel and geometrical unit, a reliable template matching algorithm based on edge shapes is proposed, and the local dynamic threshold segmentation processing is used to obtain recognition results. Simulation results show that the improved applications of the proposed algorithm can get a better result, based on the HALCON image processing software.
本文依据企业的印刷品质量检测的要求,描述了基于图像处理的目标特征识别问题的一个完整的实现和改进,改进了Canny算法进行边缘检测,同时应用了改进的可靠的模板配准算法,配以局部的动态阈值分割处理来获得识别结果,获得了准确的质量检测结果。
首先研究了图像获取的硬件系统,这是进行基于图像处理的目标特征识别研究的基础。选取设备的主要因素是设备的性能效果,同时也要仔细考虑环境因素,以及各设备之间的匹配融合,还有就是要协调考虑软件的设计,这样才能设计出最合适的系统。
其次,每幅图像都包含某种程度的噪声,综合考虑速度与效果,选用高斯滤波器来对图像进行平滑处理,对图像中的噪声进行抑制；而且应用图像分割处理技术,将图像分割为二值类型的图像,将来就可以忽略背景而直接对目标物体进行处理。在图像分割中,使用细致的、基于局部处理的动态阂值分割处理,可以得到准确的、稳定的处理结果。这样对原始图像进行了预处理,从而获得了质量较好的图像,尽力保证最终目标特征识别结果的准确性。
然后,因为如果对边缘进行处理比对整个目标物体处理的话,边缘信息的准确率远远高于整个目标物体的信息,所以需要对目标物体进行边缘提取。本文研究了Canny边缘检测算法,并在此基础上针对传统Canny算法在梯度幅值计算上对噪声过于敏感、容易检测出伪边缘的缺陷,应用了一种新的计算梯度幅值的算法,大大提高了准确度。然后针对传统Canny算法需要人为指定双阂值的缺陷,应用了一种根据非最大抑制处理的结果图像、来自动生成双阈值的方法,更全面地利用了梯度图像中的信息特征,使算法具有了更好的自适应能力,提高了自动化程度。
还有,当基于模板来检测物体时,要使图中物体的位姿与参考图像中物体的位姿相同。要进行模板配准,就是要在目标物体图像中找到模板,得到模板在图像中最合适的位姿,然后依据位姿变换关系,进行图像变换,将模板与图像中的目标物体对齐。先研究了传统的基于灰度值的模板配准算法的适用情况和使用效果,最后应用了改进的可靠的基于边缘形状的模板配准算法,配以局部的动态阂值分割处理来获得识别结果,算法准确度大大提高,也增加了对于各种质量的图像的适用范围。然后,在图像变换中,使用双线性插值变换方法,可以使得到的图像的变换结果具有平滑的外观边缘。
最后,模板与目标物体图像对齐后,质量检测算法使用的是动态阈值分割处理,设计了含有容许偏差的偏差模板,然后将与图像中的目标物体与其进行比较,以实现质量检测,结果十分可靠。
Target feature recognition is currently the focus area and hot spot in the computer vision and artificial intelligence research, which has a very wide range of applications in the military field, intelligent transportation, robot navigation, intelligent monitoring, traffic management, industrial inspection, and fingerprint identification, face recognition, iris recognition and other fields. The basic principle of the target feature recognition is that by using the target image's feature information, improving image quality with various image processing techniques, and then extracting features, finally to identify and the recognition images instead of people, based on templates images from a large number of training samples. Meanwhile, image processing technology has been paid a widespread attention to in more application fields, and has become a new potential subject with bright prospects.
With the fast development of computer technology and notice science, digital image
Processing is now developing in a deeper level. People begin the study to analyze and interpret images by using computer systems, realizing the outside world recognition with the vision systems similar to human beings'. Though having got some obvious progress in theory and Methodology, target feature recognition is still a new field needing more exportation for its complexity and high requirements,
This paper mainly studies the target feature recognition method based on the image processing. It compares the various template matching algorithms, and uses the image edge gradient vector characteristics normalized and improved. In this paper, a reliable template matching algorithm based on edge shapes is proposed, and the local dynamic threshold segmentation processing is used to obtain recognition with accurate results and high stability.
Firstly, this paper chooses gauss filter to smooth the image, rejecting the noise of the image. In the image transformation, the use of bilinear interpolation transform method can obtain the transformation results with a smooth appearance edge. In the image segmentation, the use of meticulous processing based on local processing dynamic threshold segmentation can get the exact and stable processing results. The original image preprocessed with good quality can ensure the accuracy of the ultimate target feature recognition results.
Then the Canny edge detection algorithm is studied in this paper. Because the traditional Canny algorithm for calculating the gradient amplitude is too sensitive to the noise and easily causes the false-edge detect, a new algorithm for computing gradient magnitude is proposed. With the disadvantages that the traditional Canny algorithm requires man-made dual threshold, a approach is proposed. The approach can automatically generate two-threshold according to the non-maximal inhibition image, and can make a more comprehensive use of information features of the gradient image. The proposed approach make the algorithm have better adaptive capacity, and improve the degree of automation.
At last, with the Contrastive study on the usage and effects of template matching algorithms based on the gray value, edge pixel and geometrical unit, a reliable template matching algorithm based on edge shapes is proposed, and the local dynamic threshold segmentation processing is used to obtain recognition results. Simulation results show that the improved applications of the proposed algorithm can get a better result, based on the HALCON image processing software.
引文
[1]Tanimizu, K., Meguro, S., Ishii.A High-speed defect detection method for color printed matter. Industrial Electronics Society.1990, vol.1.27-30 Nov, P:653-658
[2]Ishimaru, I, Hata, S, Hirokari, M., Color-defect classification for
printed-matter visual inspection system, Intelligent Control and Automation. 2002, vol.4.10-14 June, P:3261～3265
[3]Radke, R. J., Andre, S, Al-Kofahi,O, Roysam, B., Image change detection algorithms:a systematic survey. IEEE Transactions on Image Processing, March 2005, P:294-307.
[4]时吟,龚特.世界饮料包装的共性追求和地域特色[J].中国包装工业.2006,9(0)：49-50页.
[5]谢勇,彭涛.机器视觉及其在现代包装行业中的应用[J].株洲工学院学报.2002,16：88-89页.
[6]张强.基于机器视觉的产品检测技术研究[D].信息工程大学.2006,2
[7]孙即祥.图像处理.科学出版社,2004
[8]何斌等.Visual C++图像处理,人民邮电出版社,2002
[9]阮秋琦.数字图像处理.电子工业出版社,2003
[10]E. Ataman, V. K. Atre, Wong, A fast method for real-time median filtering, IEEE Transaction on Acoustics, Speech and Signal Processing, vol.ASSP-28, Aug,1980 P:415-421.
[11]Gonzalez C Rafael, Woods E Richard.数字图像处理(第二版).电子工业出版社,2003：463-474.
[12]李弼程,彭天强,彭波等.智能图像处理技术.电子工业出版社,2004.
[13]Stark JA. Adaptive image contrast enhancement using generalizations of histogram equalization. IEEE Trans Image Proc.2000,9(5) P:889-996.
[14]Chang Dah Chung, Wu Wenrong. Image Contrast Enhancement Based on A Histogram Transformation of Local Standard Deviation. IEEE Transactions on Medical Imaging.1998,17(4) P:518-531.
[15]张艳华,李玲远.基于GCV理论的小波自适应多阈值图像去噪方法[J].现代电子技术.2007,255(11)：62-64页.
[16]周永刚.重要阀门操作动态监测及即时报告系统的研究[D].北京化工大学.2009,7
[17]胡玉锁.一种基于鲁棒估计的自适应图像平滑方法[J].中国科学技术大学学报.2006,2
[18]吕玉增.形状仿射不变特征提取与识别研究[D].国防科学技术大学.2008,7
[19]马超杰.基于形态学和直方图分析的多源图像目标分割算法[J].航空兵器.2009,4
[20]Demigny D, Kamle T. A discrete expression of Canny's criteria for step edge detector performances evaluation. IEEE Transactions on Pattern Analysis and Machine Intelligence,1997,19(6)P:1199-1211.
[21]Demigny D. Extension of Canny's discrete criteria to second derivative filters, towards a unified approach. Proceedings of International Conference on Image Processing.1998, P:520-524.
[22]Worthington P L. Enhanced Canny edge detection using curvature consistency. In:Proceedings International Conference on Pattern Recognition. Los Alamitos, CA, USA:IEEE, Computer SOC Press,2002, P:596-599.
[23]于慧敏,廖秀秀.引入边缘信息度量的自由形变图像配准[J].浙江大学学报.2008,10：1700-1705页
[24]Xie Zhiyong, Fraine G E. Image registration using hierarchical B-splines. IEEE Transaction on Visualization and Computer Graphics.2004,10(1) P:85-94.
[25]陈明,陈武凡,冯前进,等.基于互信息量和模糊梯度相似性的医学图像配准[J].电子学报.2003,31(12)P：1835-1838页.
[26]国新毅.高速自动化灌装线上视觉检验系统研究[D].山东大学.2009,7
[27]贺顾一.复杂背景下基于视觉的动态手势识别研究[D].华东师范大学.2008,7
[28]陈德富,何通能.车牌识别的嵌入式系统研究[D].杭州：浙江工业大学.2006：9-10.
[29]董然,师卫.基于边缘检测和形态学处理的车牌定位[J].机械工程与自动化.2010,12：45-47页
[30]T. Lindeberg. Scale-Space Theory in Computer Vision. Kluwer Academic.1994
[31]宁嫒,李皖.图像去噪的几种方法分析比较[J].贵州工业大学学报(自然科学版).2005,34(4)：63-66页
[32]Chang S G, Yu Bin. Vetterli M. Spatially adaptive wavelet thresholding with context modeling for image denosing. IEEE Trans. Image Processing,2000,9 (9) P:1522-1530.
[33]Kim J Y, Kim L S, et al. An advanced contrast enhancement using partially overlapped sub block histogram equalization. IEEE Trans Circuits and Video Technology,2001,11 (4) P:475-484.
[34]李军,林宗坚.图纸扫描图像二值化阈值选取方法[J].解放军测绘学院学报.1996,13(4)：266-271页.
[35]康牧.图像处理中几个关键算法的研究[D].西安电子科技大学.2009
[36]王植,贺赛先.一种基于Canny理论的自适应边缘检测方法[J].中国图象图形学报.2004,9(8)：957-962页.
[37]L. GOTTESFELD BROWN. A.survey of image registration techniques. ACM Computing Surveys,24(4) P:325-376, December 1992.
[38]刘京京.体全息存储系统存在物像倍数误差下的像素匹配研究[D].中国优秀博硕士学位论文全文数据库(硕士).2005,(08)
[39]M. GHARAVI-ALKIIANSARI. A fast globally optimal algorithm for template matching using low-resolution pruning. IEEE Transactions on Image Processing,10(4)P:526-533, April 2001.
[40]C. STEGER. Similarity measures for occlusion, clutter, and illumination invariant object recognition. In B. RADIG, S. FLORCZYK, EDITORS, Pattern Recognition, Lecture Notes in Computer Science, Vol.2191, pp.148-154. Springer-Verlag, Berlin,2001.
[41]C. STEGER. Occlusion, clutter, and illumination invariant object recognition. In International Archives of Photogrammetry and Remote Sensing, Vol. XXXIV, part 3A,2002, P:345-350.
[42]M. ULRICH, C. STEGER. Empirical performance evaluation of object recognition methods. In H. I. CHRISTENSEN, P. J. PHILLIPS, EDITORS, Empirical Evaluation Methods in Computer Vision, P:62-76. IEEE Computer Society Press, Los Alamitos, CA,2001.
[43]梁栋,殷兵,于梅等.基于非抽样Contourlet变换的自适应阈值图像增强算法[J].电子学报.2008,36(3)：527-530页.
[2]Ishimaru, I, Hata, S, Hirokari, M., Color-defect classification for
printed-matter visual inspection system, Intelligent Control and Automation. 2002, vol.4.10-14 June, P:3261～3265
[3]Radke, R. J., Andre, S, Al-Kofahi,O, Roysam, B., Image change detection algorithms:a systematic survey. IEEE Transactions on Image Processing, March 2005, P:294-307.
[4]时吟,龚特.世界饮料包装的共性追求和地域特色[J].中国包装工业.2006,9(0)：49-50页.
[5]谢勇,彭涛.机器视觉及其在现代包装行业中的应用[J].株洲工学院学报.2002,16：88-89页.
[6]张强.基于机器视觉的产品检测技术研究[D].信息工程大学.2006,2
[7]孙即祥.图像处理.科学出版社,2004
[8]何斌等.Visual C++图像处理,人民邮电出版社,2002
[9]阮秋琦.数字图像处理.电子工业出版社,2003
[10]E. Ataman, V. K. Atre, Wong, A fast method for real-time median filtering, IEEE Transaction on Acoustics, Speech and Signal Processing, vol.ASSP-28, Aug,1980 P:415-421.
[11]Gonzalez C Rafael, Woods E Richard.数字图像处理(第二版).电子工业出版社,2003：463-474.
[12]李弼程,彭天强,彭波等.智能图像处理技术.电子工业出版社,2004.
[13]Stark JA. Adaptive image contrast enhancement using generalizations of histogram equalization. IEEE Trans Image Proc.2000,9(5) P:889-996.
[14]Chang Dah Chung, Wu Wenrong. Image Contrast Enhancement Based on A Histogram Transformation of Local Standard Deviation. IEEE Transactions on Medical Imaging.1998,17(4) P:518-531.
[15]张艳华,李玲远.基于GCV理论的小波自适应多阈值图像去噪方法[J].现代电子技术.2007,255(11)：62-64页.
[16]周永刚.重要阀门操作动态监测及即时报告系统的研究[D].北京化工大学.2009,7
[17]胡玉锁.一种基于鲁棒估计的自适应图像平滑方法[J].中国科学技术大学学报.2006,2
[18]吕玉增.形状仿射不变特征提取与识别研究[D].国防科学技术大学.2008,7
[19]马超杰.基于形态学和直方图分析的多源图像目标分割算法[J].航空兵器.2009,4
[20]Demigny D, Kamle T. A discrete expression of Canny's criteria for step edge detector performances evaluation. IEEE Transactions on Pattern Analysis and Machine Intelligence,1997,19(6)P:1199-1211.
[21]Demigny D. Extension of Canny's discrete criteria to second derivative filters, towards a unified approach. Proceedings of International Conference on Image Processing.1998, P:520-524.
[22]Worthington P L. Enhanced Canny edge detection using curvature consistency. In:Proceedings International Conference on Pattern Recognition. Los Alamitos, CA, USA:IEEE, Computer SOC Press,2002, P:596-599.
[23]于慧敏,廖秀秀.引入边缘信息度量的自由形变图像配准[J].浙江大学学报.2008,10：1700-1705页
[24]Xie Zhiyong, Fraine G E. Image registration using hierarchical B-splines. IEEE Transaction on Visualization and Computer Graphics.2004,10(1) P:85-94.
[25]陈明,陈武凡,冯前进,等.基于互信息量和模糊梯度相似性的医学图像配准[J].电子学报.2003,31(12)P：1835-1838页.
[26]国新毅.高速自动化灌装线上视觉检验系统研究[D].山东大学.2009,7
[27]贺顾一.复杂背景下基于视觉的动态手势识别研究[D].华东师范大学.2008,7
[28]陈德富,何通能.车牌识别的嵌入式系统研究[D].杭州：浙江工业大学.2006：9-10.
[29]董然,师卫.基于边缘检测和形态学处理的车牌定位[J].机械工程与自动化.2010,12：45-47页
[30]T. Lindeberg. Scale-Space Theory in Computer Vision. Kluwer Academic.1994
[31]宁嫒,李皖.图像去噪的几种方法分析比较[J].贵州工业大学学报(自然科学版).2005,34(4)：63-66页
[32]Chang S G, Yu Bin. Vetterli M. Spatially adaptive wavelet thresholding with context modeling for image denosing. IEEE Trans. Image Processing,2000,9 (9) P:1522-1530.
[33]Kim J Y, Kim L S, et al. An advanced contrast enhancement using partially overlapped sub block histogram equalization. IEEE Trans Circuits and Video Technology,2001,11 (4) P:475-484.
[34]李军,林宗坚.图纸扫描图像二值化阈值选取方法[J].解放军测绘学院学报.1996,13(4)：266-271页.
[35]康牧.图像处理中几个关键算法的研究[D].西安电子科技大学.2009
[36]王植,贺赛先.一种基于Canny理论的自适应边缘检测方法[J].中国图象图形学报.2004,9(8)：957-962页.
[37]L. GOTTESFELD BROWN. A.survey of image registration techniques. ACM Computing Surveys,24(4) P:325-376, December 1992.
[38]刘京京.体全息存储系统存在物像倍数误差下的像素匹配研究[D].中国优秀博硕士学位论文全文数据库(硕士).2005,(08)
[39]M. GHARAVI-ALKIIANSARI. A fast globally optimal algorithm for template matching using low-resolution pruning. IEEE Transactions on Image Processing,10(4)P:526-533, April 2001.
[40]C. STEGER. Similarity measures for occlusion, clutter, and illumination invariant object recognition. In B. RADIG, S. FLORCZYK, EDITORS, Pattern Recognition, Lecture Notes in Computer Science, Vol.2191, pp.148-154. Springer-Verlag, Berlin,2001.
[41]C. STEGER. Occlusion, clutter, and illumination invariant object recognition. In International Archives of Photogrammetry and Remote Sensing, Vol. XXXIV, part 3A,2002, P:345-350.
[42]M. ULRICH, C. STEGER. Empirical performance evaluation of object recognition methods. In H. I. CHRISTENSEN, P. J. PHILLIPS, EDITORS, Empirical Evaluation Methods in Computer Vision, P:62-76. IEEE Computer Society Press, Los Alamitos, CA,2001.
[43]梁栋,殷兵,于梅等.基于非抽样Contourlet变换的自适应阈值图像增强算法[J].电子学报.2008,36(3)：527-530页.