|
发布时间: 2019-06-16 |
图像分析和识别 |
|
|
收稿日期: 2018-08-09; 修回日期: 2018-12-04
基金项目: 国家重点研究发展计划项目(2017YFB0504101)
第一作者简介:
谌华, 1979年生, 男, 中国科学院国家空间科学中心博士研究生, 高级工程师, 主要研究方向为SAR图像处理。E-mail:e_chenhua@163.com;
闫敬文, 男, 教授, 博士生导师, 主要研究方向为SAR图像分析与识别、光谱压缩、小波分析及应用。E-mail:xdyanjw@126.com.
中图法分类号: TP391
文献标识码: A
文章编号: 1006-8961(2019)06-0882-08
|
摘要
目的 针对传统Grab Cut算法需要人工交互操作,无法实现合成孔径雷达(SAR)图像的自动分割,且方式单一(仅利用边界或纹理信息中的一种)的问题,提出一种综合利用边界和纹理信息的改进Grab Cut算法,实现对SAR图像目标的自动分割。方法 首先将其他格式的彩色或灰度SAR图像转化为24 bit的位图,采用图形理论对整幅SAR图像建模,根据最大流算法找到描述图的能量函数最小的割集,从而分割出目标区域;然后采用中值滤波抑制相干噪声;最后通过邻域生长算法滤除图像斑点和小目标的干扰,从而达到目标边界的连接,实现自动对SAR图像中的目标进行分割。结果 在64位Window 7环境下采用MATLAB R2014处理平台,对楼房、车库、大树、汽车群等4幅分辨率不同的SAR图像进行目标分割实验,特征目标被自动分割出来,耗时分别为1.69 s、1.58 s、1.84 s和3.09 s,相比Mean-shift和Otsu算法,平均计算效率分别提升150%和3%,并且图像中的背景杂波、目标阴影和干扰小目标均被有效去除。结论 综合利用边界和纹理信息能够有效抑制相干噪声,去除图像斑点和小目标的干扰,从而达到目标边界的连接,实现对SAR图像目标的自动分割。实验结果表明,本文算法可以满足工程化应用要求,自适应性强,分割精度高,且具有较好的鲁棒性。
关键词
合成孔径雷达图像; 目标分割; Maxflow算法; 中值滤波; 邻域生长算法
Abstract
Objective Synthetic aperture radar (SAR) systems are widely applied in many areas, such as civil and mili-tary fields, because they can operate day and night under various weather conditions. As a key and basic section of target recognition and interpretation for SAR images, SAR image segmentation has received much attention in recent years. However, SAR images suffer from strong speckle noise due to the influence of coherent illumination, which makes target segmentation in images difficult. Considering the importance of SAR image segmentation, this study proposes to segment the targets of a SAR image automatically. Many algorithms can solve SAR image target segmentation, and one of these is the GrabCut algorithm. The GrabCut algorithm, which is based on graph theory, achieves optimal segmentation and transforms the image segmentation problem into a problem of computing the maximum flow in the flow network. After this transformation, the problem can be solved with the min-cut or max-flow method. Nevertheless, the GrabCut algorithm has crucial deficiencies; for example, it not only requires artificial interaction but also merely utilizes one of the following:information, texture, or boundary information in the images. To improve such deficiencies, this study uses two kinds of information, namely, texture and boundary, for utilization in SAR images and for achieving automatic target segmentation. Method The proposed algorithm involves several steps. First, the proposed algorithm transforms a colored or gray SAR image into a 24-bit bitmap that contains substantial SAR image information. Second, with the aid of the 24-bit bitmap, a SAR image model is built according to graph theory. The model is a Gaussian mixed model that assigns each pixel in SAR images into three types of chroma spaces. Under the model framework, the energy function of the minimized description diagram is generated. Third, to segment the target region in the SAR image, the max-flow method is applied to determine the smallest cut set of the energy function in the description diagram. Coherent noise has a serious impact on image segmentation; thus, the proposed algorithm utilizes the median filter method to reduce noise in the target region and achieve precise SAR image target segmentation. Lastly, one of main problems in segmentation is that the interference of SAR image spots and small objects leads to incorrect targets during target segmentation. The neighborhood growth method, which removes specks in the SAR image target region and filters out small targets, is applied to tackle this problem and connect the target boundary. Through these steps, SAR image target segmentation can be performed automatically. Result Several state-of-the-start image segmentation algorithms, including mean-shift and Otsu segmentation algorithms, are compared with the proposed algorithm to validate its superiority. All experiments are performed in the processing platform of MATLAB R2014 in a 64-bit Windows 7 environment. In the first experiment, four different SAR images involving buildings, garages, trees, and cars are tested for image target segmentation. Results show that the proposed algorithm can segment many useful characteristics in the targets and can effectively remove background clutter, target shadows, and small interfering targets in the tested SAR images. This capability demonstrates that the proposed algorithm performs well in SAR image segmentation. A second experiment is conducted to illustrate the performance of the proposed algorithm. Here, mean-shift and Otsu segmentation algorithms and the proposed algorithm are simultaneously tested using four SAR images. As shown in the figures and tables, the proposed algorithm exhibits the best performance in SAR image target segmentation among all of the tested algorithms. The mean-shift algorithm can segment SAR image targets, but the contour boundary of the targets is fuzzy, and the computational efficiency is low. The Otsu segmentation algorithm can hardly segment targets correctly. Meanwhile, the proposed algorithm can split image targets accurately and reduce the computation time. Conclusion One of the most common methods of SAR image target segmentation is the GrabCut algorithm. However, GrabCut's precision of target segmentation is considerably affected by image background clutter and coherent noise. Meanwhile, targets in SAR images are often shielded, resulting in inaccurate image segmentation results. To address these problems, the proposed algorithm builds a Gaussian mixed model then transforms the target segmentation problem into a problem of minimizing the energy function of the description diagram. The max-flow method is used to determine the smallest cut set. The proposed algorithm can segment SAR image targets accurately through multiple iterations of the max-flow method in the SAR image color space domain and by using a median filter to remove specks in the SAR image target region and filter out small targets. However, the model in the proposed algorithm is not adaptive and thus cannot perform well in target segmentation of various SAR images. Moreover, the proposed algorithm applies the max-flow method to compute the max flow in the description diagram, thus spending considerable time in searching. In the future, we will further improve the precision of target segmentation by using an adaptive Gaussian mixed model and several useful approaches for computing the max flow to improve the performance of SAR image target segmentation.
Key words
SAR image; target segmentation; Maxflow algorithm; median filtering; neighbor-rhood growth algorithm
0 引言
合成孔径雷达(SAR)相关技术的日益进步和SAR图像分辨率的快速提高为SAR图像解译等相关研究提供了基础。其中SAR图像的分割处理技术是研究热点也是难点问题之一,它为后续目标识别中特征提取的精度提供保障,这也使得其成为SAR图像自动识别中重要的预处理步骤[1-3]。
研究表明,SAR图像感兴趣区域(ROI)分割中运用最为广泛且简单高效的算法是阈值分割算法[4]。其中使用阈值选取准则、熵决定阈值和先验信息来决定阈值的研究居多。宋文青等人[3]针对幂次变换后的图像采用一维Ostu算法进行分割,具有计算复杂度小、有利于工程应用的优点,但是幂次的选取缺乏理论支撑。吴诗婳等人[5]采用倒数灰度熵分割SAR图像中的河流目标,具有不必对条件进行初始化的优点,但是对Dirac函数以指数加权均值比率算子来替代缺乏理论上的严谨性。刘修国等人[6]对SAR图像采取超像素分割使得图像的局部边界细节处理进一步精细化。
基于图形化理论实现图像分割的核心思想是对图像进行建模,生成最小化的描述图的能量函数,然后根据最小化求得图像最大流实现图像分割。比较具有代表性的是Graph Cut[7]算法以及Rother等人[8]于2004年在Graph Cut的基础上提出的Grab Cut算法。Grab Cut算法通过非完全标注方式,首先将选中的区域标记为背景,然后对前景和背景图像分别建立高斯混合模型(GMM)[9],使用GMM参数不断地进行迭代直到达到收敛位置,从而较大程度优化了Graph Cut算法只通过一次最小估计来完成能量函数最小化的过程,因此可以更好地对图像分割。Nagahashi等人[10]将多尺度高斯平滑引入Grab Cut算法以提升算法的准确度和运算效率。周良芬等人[11]在对图像采用二次分水岭预分割的基础上,对图像的像素块采用Grab Cut分割,提高了算法的分割精度。
针对上述算法都需要人工交互操作的问题,本文提出自动分割的改进Grab Cut算法,综合利用纹理和边界信息进行自动分割。本文算法首先对SAR图像内所有像素点进行建模,采用最大流(Maxflow)算法求出最小能量函数的割集, 从而达到目标的分割;然后采用中值滤波来削弱相干噪声的干扰;最后采用邻域生长算法滤除多余的图像斑点以及少部分细小目标以期连接真实目标的边界,实现SAR目标自动分割。
1 全局Maxflow算法
Grab Cut分割是在Graph Cut的基础上,利用GMM替代灰度直方图,采用多次迭代不断更新GMM参数。本文算法首先将灰度图像统一转化为24 bit BMP彩色图像,然后使用GMM对图像的前景和背景进行建模,最后依据迭代能量最小化求得图最小割集。
1.1 SAR图像彩色数据建模
Rother等人[8]采用GMM对彩色图像进行建模是因为噪声对灰度直方图的干扰较为明显,在图像较暗时难以分辨出前景和背景信息。而基于彩色空间的图像分割,使用GMM建模的对象是图像中像素点的3个通道,受噪声干扰较小,对于图像的每个像素点用
分割过程中图的能量函数建模同Graph Cut一样,使用等价的吉布斯能量函数,可表示为分配给图像
$ E(\alpha, \boldsymbol{k}, \theta, \boldsymbol{z})=U(\alpha, \boldsymbol{k}, \theta, \boldsymbol{z})+V({\alpha}, \boldsymbol{z}) $ | (1) |
式中,
不透明代价
$ U(\alpha, \boldsymbol{k}, \theta, \boldsymbol{z})=\sum\limits_{n} D\left(\alpha_{n}, k_{n}, \theta, z_{n}\right) $ | (2) |
$ \begin{array}{c}{D\left(\alpha_{n}, k_{n}, \theta, z_{n}\right)=-\left[\log _{2} p\left(z_{n} | \alpha_{n}, k_{n}, \theta\right)+\right.} \\ {\log _{2}\left({\Pi}\left(\alpha_{n}, k_{n}\right)\right]}\end{array} $ | (3) |
式中,
$ \begin{array}{c}{D\left(\alpha_{n}, k, \theta, z_{n}\right)=\frac{1}{2} \log _{2} \operatorname{det} \Sigma\left(\alpha_{n}, k_{n}\right)-} \\ {\log _{2} \Pi\left(\alpha_{n}, k_{n}\right)+\frac{1}{2}\left[z_{n}-\mu\left(\alpha_{n}, k_{n}\right)\right]^{\mathrm{T}} \Sigma\left(\alpha_{n}, k_{n}\right)^{-1} \times} \\ {\left[z_{n}-\mu\left(\alpha_{n}, k_{n}\right)\right]}\end{array} $ | (4) |
式中,
$ \begin{array}{c}{\theta=\{\Pi(\alpha, k), \mu(\alpha, k), \Sigma(\alpha, k)} \\ {\alpha \in\{0, 1\} ; k=1, \cdots, k \}}\end{array} $ | (5) |
平滑项彩色空间中不需要计算欧氏距离,平滑项为
$ V(\alpha, z)=\gamma \sum\limits_{(m, n) \in {\boldsymbol{C}}}\left[\alpha_{n} \neq \alpha_{m}\right] \exp \left(-\beta\left\|z_{m}-z_{n}\right\|^{2}\right) $ | (6) |
式中,
$ \delta(m, n)=\left\{\begin{array}{ll}{1} & {\alpha_{m} \neq \alpha_{n}} \\ {0} & {\alpha_{m}=\alpha_{n}}\end{array}\right. $ | (7) |
常数
$ \beta=\frac{1}{2 \cdot E\left[\left(z_{m}-z_{n}\right)^{2}\right]} $ | (8) |
1.2 全局Maxflow算法
RGB彩色空间采用GMM建模后,可使用Graph Cut计算最小割集或最大流,因为Boykov等人[12]的研究证明了最大流与最小割集是等价的。Grab Cut分割算法属于半自动算法,用户要选择一定的矩形框表明其为图像背景。图像被分为背景
改进的算法中,在图像背景区域内根据最小割集(等价于最大流),采用迭代Graph Cut技术最小化,如式(1)中的能量函数得出图像的硬分割。Grab Cut算法采取用户选择区域,算法在用户区域内迭代计算找出用户选择的矩形区域内的目标,本文采用最小化局部能量函数和局部最小割集。图 2是Graph Cut算法中找到最小割集,并分割出图像中的目标区域的例子。
1.3 迭代能量最小化SAR图像分割
Grab Cut采用循环更新来求得最小的能量函数,相比于Graph Cut算法的一次迭代求得最小化能量,多次迭代的方式更加精确。此外算法还能自动更新不透明度
1) 对SAR图像中的3个区域进行初始化,
2) 初始化不透明度
3) 采用完成初始化操作的不透明度集合分别对前景和背景的GMM进行初始化;
4) 确定每个区域中像素对应的GMM参数
5) 从SAR彩色空间数据
6) 使用全局最大流算法进行目标分割,
7) 如未达到收敛阈值,则从步骤4)开始重新迭代直至阈值收敛。
2 邻域生长算法
本文的改进算法在自动分割目标后会存在部分小目标和噪声的干扰,不能将目标完全分割出来。为了去除周边细小目标及噪声干扰,本文采取邻域生长算法进行处理。邻域生长算法就是统计像素灰度值不为零的个数,如达到一定的门限阈值,则可以认定为目标,没有达到的则认定为非目标。算法先将前文分割结果中的彩色图像转化为灰度图后进行二值化,再根据邻域生长算法达到对目标区域的分割或多目标区域的提取。
对于灰度图
3 实验
所有实验均采用MATLAB R2014a,在CPU: Intel i5-3350P、内存4 GB、64位Window 7环境下完成。为了验证本文算法的有效性,分别对分辨率为3 m、5 m、5 m和10 m的楼房、车库、大树和汽车群SAR图像进行目标分割实验,同时与Mean-shift和Otsu分割算法进行对比。实验表明,本文算法能够适应不同分辨率的SAR图像,具有较好的鲁棒性。
实验结果如图 4所示,算法的运算效率分析如表 1所示。本文算法能够较好地分割出目标区域。楼房、车库、大树等SAR图像的外边缘和顶部一些特征均被分割出来,且目标轮廓比较完整,验证了改进算法的可行性。但是利用全局Maxflow分割得到的结果在部分目标区域中出现了斑点,形成伪斑点,本文采用中值滤波去除伪斑点。中值滤波器滤除这些干扰小目标的处理效果较为不错,还可以对图像目标区域起到增强的作用。采用中值滤波后,图像中仅剩下一些孤立小目标干扰,本文采用邻域生长算法对这些小目标进行抑制。同时本文算法还可以实现对高分辨率SAR图像的多目标分割,算法将汽车群中多辆汽车一并分割出来。
表 1
算法运算效率分析
Table 1
Operation efficiency analysis of algorithm
SAR图像 | 全局Maxflow/s | 中值滤波/s | 邻域生长/s | 总耗时/s |
楼房 | 1.33 | 0.04 | 0.32 | 1.69 |
车库 | 1.29 | 0.05 | 0.24 | 1.58 |
大树 | 1.56 | 0.05 | 0.23 | 1.84 |
汽车群 | 2.96 | 0.01 | 0.12 | 3.09 |
注:加粗字体表示最优结果。 |
为了验证改进算法的有效性和运算效率,分别与Mean-shift算法和Ostu算法进行对比试验,实验结果如图 5所示,算法的分割效率如表 2所示。Mean-shift算法能够大致分割出SAR图像目标轮廓,但是轮廓边界比较模糊,且算法的运算效率偏低。Otsu算法在分割出SAR图像目标轮廓的同时干扰太多,将一部分背景作为目标分割出来,分割效果较差。相比于前两种分割算法,本文改进算法能够准确地将SAR图像的轮廓分割出来,且算法的运行效率也优于其他两种算法,但是多目标分割方面运算效率稍微低于Otsu分割算法。
表 2
不同算法的分割效率
Table 2
Segmentation efficiency of different algorithms
SAR图像 | Mean-shift/s | Otsu/s | 本文算法/s |
楼房 | 4.64 | 2.26 | 1.69 |
车库 | 5.72 | 2.42 | 1.58 |
大树 | 4.41 | 2.19 | 1.84 |
汽车群 | 5.29 | 1.53 | 3.09 |
注:加粗字体表示最优结果。 |
总体来说,本文改进算法能够满足工程化应用需求,自适应性强,算法分割精度高,且具有较好的鲁棒性,可以实现对SAR图像目标的自动分割。
4 结论
SAR图像分割最常用的是恒虚警率(CFAR)检测技术。由于该算法的分割精度受背景杂波分布模型和噪声的影响较大,同时SAR图像的目标常常被遮挡,分割效果有待提升。本文提出的改进算法适用于不同分辨率的SAR图像目标分割,能精确地自适应分割图像中的目标。同时还可对不同分辨率多目标SAR图像进行分割,普适性较佳。
由于算法采用迭代能量最小化方式,使用多次迭代Graph Cut的方式求取图像彩色空间域全局最大流,算法运算量略高于Grab Cut。算法中GMM成分是随机分配的,后续将研究如何根据彩色空间的复杂度进行自适应分配
参考文献
-
[1] Wan L, You H J, Cheng Y B, et al. Research progress of synthetic aperture radar image segmentation[J]. Remote Sensing Technology and Application, 2018, 33(1): 10–24. [万玲, 尤红建, 程跃兵, 等. 合成孔径雷达图像分割研究进展[J]. 遥感技术与应用, 2018, 33(1): 10–24. ] [DOI:10.11873/j.issn.1004-0323.2018.1.0010]
-
[2] Song G L, Hou W. Overview of SAR image segmentation algorithm[J]. Computer Era, 2017(5): 1–5. [宋国磊, 侯巍. SAR图像分割算法综述[J]. 计算机时代, 2017(5): 1–5. ] [DOI:10.16644/j.cnki.cn33-1094/tp.2017.05.001]
-
[3] Song W Q, Wang Y H, Lu H X, et al. Otsu segmentation algorithm for SAR images based on power transformation[J]. Systems Engineering and Electronics, 2015, 37(7): 1504–1511. [宋文青, 王英华, 卢红喜, 等. 基于幂次变换的SAR图像Otsu分割法[J]. 系统工程与电子技术, 2015, 37(7): 1504–1511. ] [DOI:10.3969/j.issn.1001-506X.2015.07.07]
-
[4] Wu Y Q, Meng T L, Wu S H. Research progress of image thresdholding methods in recent 20 years (1994-2014)[J]. Journal of Data Acquisition and Processing, 2015, 30(1): 1–23. [吴一全, 孟天亮, 吴诗婳. 图像阈值分割方法研究进展20年(1994-2014)[J]. 数据采集与处理, 2015, 30(1): 1–23. ] [DOI:10.16337/j.1004-9037.2015.01.001]
-
[5] Wu S H, Wu Y Q, Zhou J J, et al. SAR river image segmentation based on reciprocal gray entropy and improved Chan-Vese model[J]. Acta Geodaetica et Cartographica Sinica, 2015, 44(11): 1255–1262. [吴诗婳, 吴一全, 周建江, 等. 利用倒数灰度熵和改进Chan-Vese模型进行SAR河流图像分割[J]. 测绘学报, 2015, 44(11): 1255–1262. ] [DOI:10.11947/j.AGCS.2015.20140519]
-
[6] Liu X G, Xu Q, Chen Q H, et al. Segmentation of polarimetric SAR images selectively using Wishart and K statistical description[J]. Journal of University of Electronic Science and Technology of China, 2016, 45(5): 713–719. [刘修国, 徐乔, 陈启浩, 等. 选择利用Wishart和K统计描述的极化SAR图像分割[J]. 电子科技大学学报, 2016, 45(5): 713–719. ] [DOI:10.3969/j.issn.1001-0548.2016.05.001]
-
[7] Kim J, Parra A, Li H, et al. Efficient graph-cut tattoo segmentation[C]//Proceedings of SPIE 9410, Visual Information Processing and Communication VI. San Francisco, California, United States: SPIE, 2015: #94100H.[DOI: 10.1117/12.2083419]
-
[8] Rother C, Kolmogorov V, Blake A. "GrabCut": interactive foreground extraction using iterated graph cuts[C]//Proceedings of 2004 ACM SIGGRAPH. Los Angeles, California: ACM, 2004: 309-314.[DOI: 10.1145/1186562.1015720]
-
[9] Zhang C, Cohen M, Rui Y, et al. Image segmentation using spatial-color Gaussian mixture models: US, US7885463B2[P]. 2011-02-08.
-
[10] Nagahashi T, Fujiyoshi H, Kanade T. Image segmentation using iterated graph cuts based on multi-scale smoothing[C]//Proceedings of the 8th Asian Conference on Computer Vision. Tokyo, Japan: Springer, 2007: 806-816.[DOI: 10.1007/978-3-540-76390-1_79]
-
[11] Zhou L F, He J N. Improved image segmentation algorithm based on GrabCut[J]. Journal of Computer Applications, 2013, 33(1): 49–52. [周良芬, 何建农. 基于GrabCut改进的图像分割算法[J]. 计算机应用, 2013, 33(1): 49–52. ] [DOI:10.3724/SP.J.1087.2013.00049]
-
[12] Boykov Y, Kolmogorov V. An experimental comparison of min-cut/max-flow algorithms for energy minimization in vision[C]//Proceedings of the 3rd International Workshop on Energy Minimization Methods in Computer Vision and Pattern Recognition. Sophia Antipolis, France: Springer, 2001: 359-374.[DOI: 10.1007/3-540-44745-8_24]