基于形态学和小波变换的弱小目标检测
|
摘要
复杂背景下弱小目标的检测在监视和预警系统中有着十分重要的作用,也是当前国内外研究的热点。由于目标距离观察点较远,其成像大小只有几到十几个像素,且目标淹没于背景杂波干扰之中,给人们的检测带来了很大的困难和挑战。
本文首先分析研究了复杂背景下弱小目标的特性,包括灰度特征和运动特征,这是进行所有检测工作的基础。其中,弱小目标的灰度特征主要表现为灰度曲面上的微小“凸起”,则检测出这些微小“凸起”是弱小目标单帧检测的主要任务;而运动特征主要表现为目标运动的独立性和运动轨迹的连续性,则检测弱小目标的独立运动和连续轨迹是弱小目标运动检测的主要任务。据此,本文设计了弱小目标的检测系统,主要包括目标图像预处理、目标分割、目标跟踪与检测等。
本文的研究重点是弱小目标图像的预处理算法,其重要作用是增强目标、抑制噪声,提高目标的可检测性,同时还可以减少后续工作的计算量,提高检测效率。本文重点研究了基于形态学和小波变换的两种预处理算法。其中,基于形态学的预处理算法,采用Top-hat变换来抑制背景图像,增强目标。在形态学算法中,结构元的选择是一个重要的内容,本文根据弱小目标的灰度特征,选取了“盘形”结构元。另外,基于小波变换的预处理算法,重点分析了低频子带置零法和高频子带相乘消噪法。在小波分析中,小波基的选择是一个重要的内容,本文针对小波分析在图像处理和目标检测中的应用,选择了双正交的小波基。
为了有效地提取目标,并进一步减少后续工作的运算量,在弱小目标分割中,充分考虑不同目标图像自身的特点,引入了基于目标图像均值和方差的自适应局部阈值分割方法。
以上工作都是在单帧图像中进行的,主要利用了弱小目标的灰度特征,也就是静态特征。但是,仅利用目标的灰度信息还不能有效捕获真实的目标,因为还存在很多虚警,例如类目标干扰。因此,还需要在多帧序列图像中,利用弱小目标的运动特征,即目标运动轨迹的连续性,进一步抑制虚警,从而捕获真正的目标,并获得其运动轨迹。
Detection of dim small moving targets in sequence images with complex clutter background is of vital importance to modern surveillance and alarm system, and it's hot all over the world.For long distance, the dim small moving target is usually composed of several pixels without specific shape and texture.And usually, the target is emerged in background clutter, which causes a lot of difficulties and challenges in detection.
Firstly,the intensity features and motion features of the dim small moving target with complex clutter background are studied in this dissertation,and it's the basis for target detection.From the study on intensity features of the target, the dim small target in scene image presents as some special gray bubble, and the detection of the special bubble is the key task for target detection in a single image.The study on motion features of the target shows that, the motion of the moving target in sequence images has their independent motion and the track of the target is continuous. The detection of the independent motion and continuous track is the key task for motion detection of the target. On these grounds, a detecting system is designed in this dissertation,which includes target image pre-processing,target segmentation and target detection.
This dissertation focuses on stduying the pre-processing algorithms of target image. The pre-processing algorithm is important in detecting target by enhancing the target and supressing noises,and it also can reduce the calculation and enhance the efficiency of detection. Two pre-processing algorithms about morphology and wavelet transform are mainly studied in this dissertation.For morphology pre-processing,the top-hat transform is adopted to enhance the target and remove the background.The selection of proper structure element is vital in the morphology,so a disk shape structure element is selected according to the intensity features of dim small target.And in wavelet transform pre-processing,the method of letting low frequency subband to zero and the method of letting high frequency subband multiply are mainly analyzed. In wavelet transform,the selection of base wavelet is important,so the biorthogonal wavelet is selected according to the image processing and target detection.
In order to extract suspicious targets effectively and reduce the calculation,the adaptive threshold based on statistical mean and standard of image is adopted in target segmentation,and this threshold can adapt to more images with complex background.
All the above work is just for a single frame image,and the intensity information of dim small targets is mainly used.But the real target can’t be captured only by using the intensity features,because some false alarms exist,especially the target-like interfence. Therefor,the motion features of dim small targets in sequence images need to be used.So the continuity of track can be used to suppress the false alarms,then the real target can be captured and the track can be obtained.
本文首先分析研究了复杂背景下弱小目标的特性,包括灰度特征和运动特征,这是进行所有检测工作的基础。其中,弱小目标的灰度特征主要表现为灰度曲面上的微小“凸起”,则检测出这些微小“凸起”是弱小目标单帧检测的主要任务;而运动特征主要表现为目标运动的独立性和运动轨迹的连续性,则检测弱小目标的独立运动和连续轨迹是弱小目标运动检测的主要任务。据此,本文设计了弱小目标的检测系统,主要包括目标图像预处理、目标分割、目标跟踪与检测等。
本文的研究重点是弱小目标图像的预处理算法,其重要作用是增强目标、抑制噪声,提高目标的可检测性,同时还可以减少后续工作的计算量,提高检测效率。本文重点研究了基于形态学和小波变换的两种预处理算法。其中,基于形态学的预处理算法,采用Top-hat变换来抑制背景图像,增强目标。在形态学算法中,结构元的选择是一个重要的内容,本文根据弱小目标的灰度特征,选取了“盘形”结构元。另外,基于小波变换的预处理算法,重点分析了低频子带置零法和高频子带相乘消噪法。在小波分析中,小波基的选择是一个重要的内容,本文针对小波分析在图像处理和目标检测中的应用,选择了双正交的小波基。
为了有效地提取目标,并进一步减少后续工作的运算量,在弱小目标分割中,充分考虑不同目标图像自身的特点,引入了基于目标图像均值和方差的自适应局部阈值分割方法。
以上工作都是在单帧图像中进行的,主要利用了弱小目标的灰度特征,也就是静态特征。但是,仅利用目标的灰度信息还不能有效捕获真实的目标,因为还存在很多虚警,例如类目标干扰。因此,还需要在多帧序列图像中,利用弱小目标的运动特征,即目标运动轨迹的连续性,进一步抑制虚警,从而捕获真正的目标,并获得其运动轨迹。
Detection of dim small moving targets in sequence images with complex clutter background is of vital importance to modern surveillance and alarm system, and it's hot all over the world.For long distance, the dim small moving target is usually composed of several pixels without specific shape and texture.And usually, the target is emerged in background clutter, which causes a lot of difficulties and challenges in detection.
Firstly,the intensity features and motion features of the dim small moving target with complex clutter background are studied in this dissertation,and it's the basis for target detection.From the study on intensity features of the target, the dim small target in scene image presents as some special gray bubble, and the detection of the special bubble is the key task for target detection in a single image.The study on motion features of the target shows that, the motion of the moving target in sequence images has their independent motion and the track of the target is continuous. The detection of the independent motion and continuous track is the key task for motion detection of the target. On these grounds, a detecting system is designed in this dissertation,which includes target image pre-processing,target segmentation and target detection.
This dissertation focuses on stduying the pre-processing algorithms of target image. The pre-processing algorithm is important in detecting target by enhancing the target and supressing noises,and it also can reduce the calculation and enhance the efficiency of detection. Two pre-processing algorithms about morphology and wavelet transform are mainly studied in this dissertation.For morphology pre-processing,the top-hat transform is adopted to enhance the target and remove the background.The selection of proper structure element is vital in the morphology,so a disk shape structure element is selected according to the intensity features of dim small target.And in wavelet transform pre-processing,the method of letting low frequency subband to zero and the method of letting high frequency subband multiply are mainly analyzed. In wavelet transform,the selection of base wavelet is important,so the biorthogonal wavelet is selected according to the image processing and target detection.
In order to extract suspicious targets effectively and reduce the calculation,the adaptive threshold based on statistical mean and standard of image is adopted in target segmentation,and this threshold can adapt to more images with complex background.
All the above work is just for a single frame image,and the intensity information of dim small targets is mainly used.But the real target can’t be captured only by using the intensity features,because some false alarms exist,especially the target-like interfence. Therefor,the motion features of dim small targets in sequence images need to be used.So the continuity of track can be used to suppress the false alarms,then the real target can be captured and the track can be obtained.
引文
[1]杨卫平,沈振康.红外图像序列小目标检测预处理技术.红外与激光工程,No.2,1998,23-28
[2] Barnett J. Statistical analysis of median subtraction filtering with application to point target detection in infrared background, SPIE, Vol.1050, 1989,10-18
[3]徐英.红外图像序列中运动弱小目标检测方法研究.红外技术,Vol.24, No.6, 2002,27-30
[4] Muller M. Saliency measures in clutter IR image for ATR, SPIE, Vol.3699,150-154
[5] Scharfand L. L. Friedlander B. Matched subspace detectors, IEEE Trans. On Sig. Proc. Aug. 1994,42(8),2146-2156
[6] Youn, Hyoung-Sun and Chen, Chi-Chih Automatic G.P.R. target detectiion and clutter reduction using neural network, SPIE, Vol.4758, 2002,579-582
[7] Zhu Zhenfu, Grayscale morphological filter for small target detection, SPIE, 2000, 28-34
[8] Zhu, Z., Morphological filter and it's-applications in TV tracker, SPIE, VOl.4170, 2001,265-271
[9] Deshpande S. D., Er M. H., Ronda V., Max-mean and max-median filters for detection of small targets, SPIE, Vol.3809,74-83
[10]黄林梅.低信噪比红外弱小运动目标先跟踪后检测的检测方法研究.华中科技大学硕士学位论文,2004,35-41
[11] Liou R. and Azimi-Sadjadi M. R., Dim target detection using high order correlation method, IEEE Trans. On Aerospace and Electronic System,29(1),1994,197-212
[12]陈朝阳,张桂林.红外警戒系统小目标实时检测方法.红外与毫米波学报,1998,Vol.17,No.4:12-15
[13]候格贤,吴成柯.基于遗传算法的目标检测定位方法.系统工程与电子技术,1997,Vol.18,No.6:17-20
[14]徐军.红外图像中弱小目标检测技术研究.西安电子科技大学博士学位论文,2001,34-39
[15]程德杰.弱小运动图像目标形态检测理论与技术研究.电子科技大学博士学位论文,2006
[16]崔屹.图像处理与分析:数学形态学方法及应用.科学出版社,2002,72-81
[17] Zhang Y. J., Segmentation evaluation and comparison: a study of several algorithm, SPIE, Vol.2094, 1993,217-226
[18] Zeng Ming and Li Jianxun.The design of Top-Hat morphological filter and application to infrared target detection, Infrared Physics and Technology, Vol.48, No.1, April, 2006,45-51
[19]李世雄.小波变换及其应用.大连:高等教育出版社,1997.7,44-52
[20]陈仲英,巫斌.小波分析.北京:科学出版社,2007.8,162-175
[21]李建平.小波十讲.北京:国防工业出版社,2004.5,77-82
[22]孙廷奎.小波分析及其应用.北京:机械工业出版社,2005.3,82-96
[23]董长虹.Matlab小波分析工具箱原理与应用.北京:国防工业出版社,2004.1,115-124
[24]方浩宇.序列图像中的目标检测与跟踪算法研究.电子科技大学硕士学位论文,2006,25-32
[25]刘钢.基于小波变换的航空图像处理及动载体多目标跟踪方法研究,中国科学院研究生博士学位论文,2005,47-55
[26]李庆华.复杂背景条件下的红外图像中小目标的检测.哈尔滨工程大学硕士学位论文,2006,31-38
[27]李利荣.基于小波变换的小目标检测方法研究.华中科技大学硕士学位论文,2004,35-42
[28]吕宗祺.红外点目标探测的信号处理算法及实现.航空电子技术,1995,12-16
[29]郝晓冉.序列图像中运动点目标检测方法的研究.山东大学硕士学位论文,2005,44-51
[30]张鹏.弱小点目标的检测.上海交通大学硕士学位论文,2007,25-34
[31]史凌峰.红外弱小目标检测方法研究.西安电子科技大学硕士学位论文,2003,22-27
[32]贺鹏.Hough变换航迹起始算法研究.西北工业大学硕士论文,2007,18-26
[2] Barnett J. Statistical analysis of median subtraction filtering with application to point target detection in infrared background, SPIE, Vol.1050, 1989,10-18
[3]徐英.红外图像序列中运动弱小目标检测方法研究.红外技术,Vol.24, No.6, 2002,27-30
[4] Muller M. Saliency measures in clutter IR image for ATR, SPIE, Vol.3699,150-154
[5] Scharfand L. L. Friedlander B. Matched subspace detectors, IEEE Trans. On Sig. Proc. Aug. 1994,42(8),2146-2156
[6] Youn, Hyoung-Sun and Chen, Chi-Chih Automatic G.P.R. target detectiion and clutter reduction using neural network, SPIE, Vol.4758, 2002,579-582
[7] Zhu Zhenfu, Grayscale morphological filter for small target detection, SPIE, 2000, 28-34
[8] Zhu, Z., Morphological filter and it's-applications in TV tracker, SPIE, VOl.4170, 2001,265-271
[9] Deshpande S. D., Er M. H., Ronda V., Max-mean and max-median filters for detection of small targets, SPIE, Vol.3809,74-83
[10]黄林梅.低信噪比红外弱小运动目标先跟踪后检测的检测方法研究.华中科技大学硕士学位论文,2004,35-41
[11] Liou R. and Azimi-Sadjadi M. R., Dim target detection using high order correlation method, IEEE Trans. On Aerospace and Electronic System,29(1),1994,197-212
[12]陈朝阳,张桂林.红外警戒系统小目标实时检测方法.红外与毫米波学报,1998,Vol.17,No.4:12-15
[13]候格贤,吴成柯.基于遗传算法的目标检测定位方法.系统工程与电子技术,1997,Vol.18,No.6:17-20
[14]徐军.红外图像中弱小目标检测技术研究.西安电子科技大学博士学位论文,2001,34-39
[15]程德杰.弱小运动图像目标形态检测理论与技术研究.电子科技大学博士学位论文,2006
[16]崔屹.图像处理与分析:数学形态学方法及应用.科学出版社,2002,72-81
[17] Zhang Y. J., Segmentation evaluation and comparison: a study of several algorithm, SPIE, Vol.2094, 1993,217-226
[18] Zeng Ming and Li Jianxun.The design of Top-Hat morphological filter and application to infrared target detection, Infrared Physics and Technology, Vol.48, No.1, April, 2006,45-51
[19]李世雄.小波变换及其应用.大连:高等教育出版社,1997.7,44-52
[20]陈仲英,巫斌.小波分析.北京:科学出版社,2007.8,162-175
[21]李建平.小波十讲.北京:国防工业出版社,2004.5,77-82
[22]孙廷奎.小波分析及其应用.北京:机械工业出版社,2005.3,82-96
[23]董长虹.Matlab小波分析工具箱原理与应用.北京:国防工业出版社,2004.1,115-124
[24]方浩宇.序列图像中的目标检测与跟踪算法研究.电子科技大学硕士学位论文,2006,25-32
[25]刘钢.基于小波变换的航空图像处理及动载体多目标跟踪方法研究,中国科学院研究生博士学位论文,2005,47-55
[26]李庆华.复杂背景条件下的红外图像中小目标的检测.哈尔滨工程大学硕士学位论文,2006,31-38
[27]李利荣.基于小波变换的小目标检测方法研究.华中科技大学硕士学位论文,2004,35-42
[28]吕宗祺.红外点目标探测的信号处理算法及实现.航空电子技术,1995,12-16
[29]郝晓冉.序列图像中运动点目标检测方法的研究.山东大学硕士学位论文,2005,44-51
[30]张鹏.弱小点目标的检测.上海交通大学硕士学位论文,2007,25-34
[31]史凌峰.红外弱小目标检测方法研究.西安电子科技大学硕士学位论文,2003,22-27
[32]贺鹏.Hough变换航迹起始算法研究.西北工业大学硕士论文,2007,18-26