-
选择性注意机制是视觉系统的重要特征,这种机制可以将视场中的信息按照关键性排序的方式进行高速处理,获取重要局部区域,使得人类可以迅速将注意力集中至该区域 [ 15- 16] 。这些区域的特征即具备较高的对比度。人类的这种视觉特性对计算机视觉而言启发较大,可有效剔除无效信息,从而将计算资源集中至特征明显区域。
受生物的注意力机制启发,一些学者提出将图像局部领域之间的对比度作为特征,再从全局出发进行局部区域特征检测的方法,即显著度方法。当前基于显著度的视觉算法主要有频谱残差方法、视觉注意方法等。
视觉显著度的方法如 图1所示:首先定义3×3的正方形块,每个块的大小都为 m× mpixel,中心块为M,周边块从M1至M8。
基于生物视觉的研究成果,采用对比度来衡量生物对目标识别的特征,对于图像块的像素中心坐标为( a, b)的点,定义与第 c个领域块的对比度为:
$${D}_c^{\left( {a,b} \right)} = \frac{{{{\left[ {\max \left( {I_M^{\left( {a,b} \right)}} \right)} \right]}^2}}}{{{\rm{mean}}\left[ {I_{{M_c}}^{\left( {a,b} \right)}} \right]}}$$ 式中:
$\max \left( {I_M^{\left( {a,b} \right)}} \right)$ 为中心像素块M中像素灰度值最大值;${\rm{mean}}\left[ {I_{{M_c}}^{\left( {a,b} \right)}} \right]$ 为第 c个领域图像块中像素灰度值的平均值。定义像素中心坐标为( a, b)处的领域最大对比度为:
$${D}^{\left( {a,b} \right)} = \mathop {\max }\limits_{c = 1,2,...,8} \left\{ {\frac{{{{\left[ {\max \left( {I_M^{\left( {a,b} \right)}} \right)} \right]}^2}}}{{{\rm{mean}}\left[ {I_{{M_c}}^{\left( {a,b} \right)}} \right]}}} \right\}$$ 即对于8个领域图像块进行遍历,寻找最大对比度的值,作为该点( a, b)的领域最大对比度值。
通过全图计算最大领域对比度值,可获得感兴趣区域,为进一步寻找目标极大缩小搜索范围。
图像的感兴趣区域的显著图提取方法按照二值图的分割方法完成,计算公式如下:
$${Th} = {\bar{D}} + {k}{\sigma _{{D}}}$$ 式中:
${\bar{D}}$ 为最大领域对比度值的均值;${\sigma _{D}}$ 为最大领域对比度值的离散度(1 σ)。 -
基于最大对比度算法的检测性能在图像滑动块位于红外图像的背景变化剧烈,当目标正好位于该处时,检测性能较差。这是因为最大对比度算法在背景剧烈变化处(比如海天交界处)时,在凸显目标的同时也放大了边缘纹理的对比度,导致目标被周围图像信息干扰,有时甚至提取不出含目标的图像区域块。另外,提取的感兴趣区域面积过大,也会导致图像处理算法实时性较差。
从上节公式出发,类似可得:
$${{{D}}^{\left( {a,b} \right)}} = \mathop {{\rm{median}}}\limits_{c = 1,2,...,8} \left\{ {\frac{{{{\left[ {\max \left( {I_M^{\left( {a,b} \right)}} \right)} \right]}^2}}}{{{\rm{mean}}\left[ {I_{{M_c}}^{\left( {a,b} \right)}} \right]}}} \right\}$$ 其中,求最大对比度的计算过程被求中值对比度的计算过程取代,这样做的好处是当滑动图像块在边缘扫描时,对比度不显著。另外,当目标被扫描时,可以获得更高的对比度。在完成基于最恰对比度的过程后,获取全图的感兴趣区域的方法依然是阈值分割方法。
下面对不同类型区域的对比度解算过程进行分析:
(1)图像滑动块包含目标时,背景位于起伏较小的区域
如 图2(a)右侧滑动图像块,领域块为平滑的背景,因此对领域块求取的8个对比度值差异很小,于是,求中值后得到的结果与最大对比度算法得到的结果基本一致,目标感兴趣区域块的提取不受影响。
图 2 基于最恰对比度的目标特征区域提取算法分析图
Figure 2. Analysis chart of target feature region extraction algorithms based on optimal contrast
(2)图像滑动块不包含目标时,背景位于起伏较小的区域
如 图2(a)左侧滑动图像块,依然为平滑的领域块,对领域块求取的8个对比度值差别依然不大,求中值后的结果与最大对比度算法也保持一致,不会增强背景对比度值。
(3)图像滑动块包含目标时,背景位于图像边缘区域
如 图2(b)左侧滑动图像块,采用传统最大对比度值的计算结果会增强边缘对比度,而采用最恰对比度算法在抑制边缘对比度的同时,可以增强目标。
(4)图像滑动块不包含目标时,背景位于图像边缘区域。
如 图2(b)右侧滑动图像块,采用文中提出的最恰对比度算法处理得到的边缘被极大抑制。在后续实验部分会继续验证该结论。
算法流程如 图3所示。
-
文中提出的算法在笔记本电脑上实现,电脑配置为:Intel Core i5,主频2.50 GHz,内存为4.00 GB,32位操作系统。操作系统为Windows 7,开发软件为Matlab2012。
实验采用18帧存在弱小红外目标的图像进行有效性检测分析,其中,地面背景图像1帧,海天交界背景5帧,海洋背景1帧,天空背景11帧。滑动图像块大小为12×12。采用局部均值信噪比、局部均值信噪比增益和背景抑制因子衡量算法 [ 17] 对红外弱小目标的检测能力。计算公式为:
$$ \begin{array}{l} {{{P}}_{{\rm{SNR}}}}{\rm{ = mean}}\left( {{{{I}}_{{\rm{Local}}}}{\rm{ - }}{{{\bar{ I}}}_{{\rm{Local}}}}} \right)/{\sigma _{{\rm{Local}}}}\\ {\rm{GAI}}{{\rm{N}}_{{\rm{SNR}}}}{\rm{ = }}{{{P}}_{{\rm{SNR}}}}/{{{P}}_{{\rm{SNR0}}}}\\ {\rm{BSF = }}{{{N}}_{{\rm{before}}}}{\rm{/}}{{{N}}_{{\rm{after}}}} \end{array}$$ 式中:mean(·)为求均值的函数;
${{{I}}_{{\rm{Local}}}}$ 为算法处理后局部区域的图像函数;${{\bar{I}}_{{\rm{Local}}}}$ 为算法处理后局部区域的图像均值;${\sigma _{{\rm{Local}}}}$ 为算法处理后局部区域的图像灰度离散度;${{{P}}_{{\rm{SNR}}}}$ 为算法处理后局部区域信噪比值;${{{P}}_{{\rm{SNR0}}}}$ 为算法处理前局部区域信噪比值;${\rm{GAI}}{{\rm{N}}_{{\rm{SNR}}}}$ 为局部均值信噪比增益;BSF为背景抑制因子,滤波前的噪声比滤波后的噪声。从实验结果如 表1~ 4和 图4~ 图7所示,文中提出的方法比中值滤波方法的效果好很多。与参考文献[ 18]提出的方法相比,在大部分情况下效果更好。通过观察3种方法的二值图可知,文中提出方法的检测虚警率极低,而中值滤波方法及文献提出的方法虚警率较高,对目标检测效果较差。从4组实验的三维灰度曲线可知,目标基本湮灭在起伏的背景中,因此,目标比较难以检测,3种方法抑制背景噪声的能力相当,但文中提出的方法对目标的增强效果显著,有利于检测目标。
表 1 实验结果1
Table 1. Experimental results 1
Picture 1 SCR SCRgain BSF Median filter method 7.5 5.7 3 Ref. [ 18] method 6.9 5.3 2 Proposed method 10.2 7.8 2.7 表 2 实验结果2
Table 2. Experimental results 2
Picture 1 SCR SCRgain BSF Median filter method 13.1 9.8 2.9 Ref. [ 18] method 9.0 6.8 3.5 Proposed method 13.6 10.3 2.8 表 3 实验结果3
Table 3. Experimental results 3
Picture 1 SCR SCRgain BSF Median filter method 6.2 4.7 2.4 Ref. [ 18] method 8.1 6.2 2.6 Proposed method 9 6.9 2.6 表 4 实验结果4
Table 4. Experimental results 4
Picture 1 SCR SCRgain BSF Median filter method 9.9 3.2 2 Ref. [ 18] method 6.2 2 1.8 Proposed method 11 3.5 1.8 由 图4~ 图7可知,该方法用于导弹末制导及红外侦察预警中,应用效果较好,可在海面、天空背景下有效检测弱小目标,提高末制导的目标检测概率,提升目标检测的侦察预警距离,为告警拦截系统争取反应时间。
Detection of dim and small infrared targets based on the most appropriate contrast saliency analysis
-
摘要: 针对红外图像弱目标检测困难的现状,提出一种基于最恰对比度显著性分析的红外弱小目标检测方法,在滑动窗口中采用了非线性处理技术对图像进行处理,避免了传统的显著度分析算法处理图像时在景像边缘处产生的显著度值干扰问题,同时不影响在目标区域对目标的提取能力。进行了大量的半实物仿真实验,结果表明,虽然提出的方法在背景抑制因子中未明显提高,但在均值信噪比和信噪比增益两个指标中对目标检测性能明显增强。在图像处理后的三种方法视觉对比图中,效果最好。Abstract: Aiming at the current situation of dim target detection in infrared image, a dim and small target detection method based on the most appropriate contrast saliency analysis was proposed. In sliding serial port, the non-linear processing technology was used to process the image, which avoided the saliency produced by traditional saliency analysis algorithm when processing the image at the scene edge. The problem of value interference does not affect the ability of target extraction in the target area. A large number of hardware-in-the-loop simulation experiments were carried out. The results show that although the proposed method can not improve significantly in the background suppression factor, the performance of target detection in the two indicators of mean signal-to-noise ratio and signal-to-noise ratio gain of the proposed method are significantly enhanced. And among the three methods of image processing, the effect is the best.
-
表 1 实验结果1
Table 1. Experimental results 1
Picture 1 SCR SCRgain BSF Median filter method 7.5 5.7 3 Ref. [ 18] method 6.9 5.3 2 Proposed method 10.2 7.8 2.7 表 2 实验结果2
Table 2. Experimental results 2
Picture 1 SCR SCRgain BSF Median filter method 13.1 9.8 2.9 Ref. [ 18] method 9.0 6.8 3.5 Proposed method 13.6 10.3 2.8 表 3 实验结果3
Table 3. Experimental results 3
Picture 1 SCR SCRgain BSF Median filter method 6.2 4.7 2.4 Ref. [ 18] method 8.1 6.2 2.6 Proposed method 9 6.9 2.6 表 4 实验结果4
Table 4. Experimental results 4
Picture 1 SCR SCRgain BSF Median filter method 9.9 3.2 2 Ref. [ 18] method 6.2 2 1.8 Proposed method 11 3.5 1.8 -
[1] Hou Wang, Mei Fenghua. Small target detection method based on maximum likelihood estimation [J]. Optics & Optoelectronic Technology, 2019, 17(1): 42-50. (in Chinese) [2] Zhong Lei, Li Yong, Mou Zhiying, et al. Detection method for weak target under unknown strong clutter based on DP-TBD [J]. Systems Engineering and Electronics, 2019, 41(1): 43-49. (in Chinese) [3] Deng Jianxun, Xiong Zhongyang, Deng Xin. Dim target detection in air based on information fusion [J]. Electronics Optics & Control, 2018, 25(2): 5-10. (in Chinese) doi: 10.3969/j.issn.1671-637X.2018.02.001 [4] Wang Jingyu, Wang Xianyu, Zhang Ke, et al. Small UAV target detection model based on deep neural network [J]. Journal of Northwestern Polytechnical University, 2018, 36(2): 258-263. (in Chinese) doi: 10.3969/j.issn.1000-2758.2018.02.008 [5] Wang Zhi, Chen Ping, Pan Jinxiao. Object detection based on deep learning in complex background [J]. Journal of Chongqing University of Technology(Natural Science), 2018, 32(4): 171-176. (in Chinese) [6] Tian Wen, Li Suruo. Infrared dim target detection algorithm based on frequency-space domain mapping and multi-scale top-hat transform [J]. Optical Technology, 2018, 44(3): 325-332. (in Chinese) [7] Miao Xiaokong, Wang Chunping. Single frame infrared (IR) dim small target detection based on improved sobel operator [J]. Opto-Electronic Engineering, 2016, 43(12): 119-125. (in Chinese) doi: 10.3969/j.issn.1003-501X.2016.12.019 [8] Shi Xiaogang, Bai Xiaodong, Li Lijuan, et al. A dual-band infrared dim target detection algorithm based on wavelet analysis [J]. Infrared Technology, 2016, 38(8): 688-692. (in Chinese) [9] Song Qiang. Radar weak multiple-target detection and tracking method under clutter interference [J]. Journal of CAEIT, 2018, 13(5): 601-605. (in Chinese) doi: 10.3969/j.issn.1673-5692.2018.05.019 [10] Zhao Dong, Zhou Huixin, Qin Hanlin, et al. Infrared dim-small target tracking based on guided image filtering and Kernelized correlation filtering [J]. Acta Optica Sinica, 2018, 38(2): 204001. (in Chinese) doi: 10.3788/AOS201838.0204001 [11] Sun Ting, Geng Guohua, Zheng Enming, e tc. The weak target detection method based on the subspace bearing stability [J]. Journal of Shanghai Jiaotong University, 2018, 52(4): 480-487. (in Chinese) [12] Shuai Xiaofei, Zhu Yujun, Zhan Xu. Weak target accumulation detection based on keystone transform and implementation methods in engineering [J]. Fire Control Radar Technology, 2018, 47(3): 26-30. (in Chinese) doi: 10.3969/j.issn.1008-8652.2018.03.007 [13] Wang Yuanbin, Yin Yang. Research on the insulation equipment fault detection based on infrared weak target extraction [J]. Infrared Technology, 2018, 40(2): 193-199. (in Chinese) [14] Li Pengfei, Ai Yonghong, Jing Junfeng, et al. Clothing pieces stripes alignment method based on image matching [J]. Journal of Electronic Measurement and Instrumentation, 2016, 30(12): 1982-1991. (in Chinese) [15] Di Nan, Li Guiju, Chen Chunning, et al. Image matching technology research based on normalized DOG features [J]. Journal of Electronic Measurement and Instrument, 2014, 28(6): 585-590. (in Chinese) [16] Jiang Jianguo, An Hongxin, Qi Meibin, et al. Ihpid target detection algorithm in complex scene [J]. Journal of Electronic Measurement and Instrument, 2012, 26(3): 261-266. (in Chinese) [17] He Yujie, Li Min, Zhang Jinli, et al. Infrared small target detection method based on correlation filter [J]. Acta Optica Sinica, 2016, 36(5): 0512001. (in Chinese) doi: 10.3788/AOS201636.0512001 [18] Wang Gang, Chen Yongguang, Yang Suochang, et al. Detection of infrared dim small target based on image patch contrast [J]. Optics and Precision Engineering, 2015, 23(5): 1424-1433. (in Chinese) doi: 10.3788/OPE.20152305.1424