﻿ 基于梯度比率的SAR图像局部特征提取方法研究
«上一篇
 文章快速检索 高级检索

 智能系统学报  2017, Vol. 12 Issue (3): 286-292  DOI: 10.11992/tis.201603025 0

### 引用本文

WANG Qing, TANG Tao, XIANG Deliang, et al. Research on local feature extraction of SAR images based on gradient ratio[J]. CAAI Transactions on Intelligent Systems, 2017, 12(3): 286-292. DOI: 10.11992/tis.201603025.

### 文章历史

Research on local feature extraction of SAR images based on gradient ratio
WANG Qing, TANG Tao, XIANG Deliang, SU Yi
School of Electronic Science and Engineering, National University of Defense Technology, Changsha 410073, China
Abstract: In this study, we investigate a local binary pattern (LBP) operator based on a difference calculation and a local gradient ratio pattern (LGRP) operator based on a gradient ratio. First, we introduce a basic and several other LBP operators and evaluate the performance of the LBP operators using optical image and synthetic aperture radar (SAR) image analysis. To address the problem of LBP's sensitivity to multiplicative noise in SAR images, we use the LGRP calculator based on the gradient ratio, combined with the anti-rotation characteristics of a rotation-invariant LBP, and propose an improved rotation-invariant LGRP characteristic for SAR images. Our experimental results demonstrate that the proposed feature has good invariant performance in target recognition and image texture slice matching with changes in the angle of attitude.
Key words: SAR image    feature extraction    local binary pattern    gradient ratio    rotation-invariant

SAR固有的相干成像方式会导致描述同一目标场景的多幅图像之间出现几何和辐射差异。图像匹配通过将两幅图像的相似性进行比较，根据比较结果快速地进行SAR图像识别，成为进一步挖掘目标场景信息变化的前提[1]

1 局部二值模式 1.1 局部二值模式的数学基础

 图 1 LBP算子计算过程 Fig.1 Calculation process of the original LBP
 ${\rm{LB}}{{\rm{P}}_{P, R}} = \sum\limits_{P = 0}^{P-1} {s\left( {{g_{\rm{p}}}-{g_{\rm{c}}}} \right){2^P}}$

1.2 旋转不变局部二值模式

 ${\rm{LBP}}_{P, R}^{ri} = \min \left\{ {{\rm{ROR}}\left( {{\rm{LB}}{{\rm{P}}_{P, R}}, i} \right)\left| {i = 0, 1, \cdots, P-1} \right.} \right\}$

1.3 光学与SAR图像中LBP算子的抗旋转性比较分析

 图 2 图像旋转前LBP和旋转不变LBP算子比较 Fig.2 Comparison of LBP and rotation invariant LBP operator before image rotation
 图 3 图像旋转后LBP和旋转不变LBP算子比较 Fig.3 Comparison of LBP and rotation invariant LBP operator after image rotation
1.3.1 光学图像实验

1.3.2 SAR图像实验

 图 4 T72旋转前LBP和旋转不变LBP对比实验 Fig.4 Comparison of LBP and rotation invariant LBP operator before T72 image rotation
 图 5 T72旋转后LBP和旋转不变LBP对比实验 Fig.5 Comparison of LBP and rotation invariant LBP operator after T72 image rotation

 图 6 BTR70系列图像实验 Fig.6 Experiments on BTR70 series image
 图 7 BMP2系列图像实验 Fig.7 Experiments on BMP2 series image

2 改进的LGRP特征提取方法 2.1 局部梯度比率二值模式

LBP算法易受乘性相干斑噪声的影响，对图像局部梯度变化不敏感，例如在边缘、角点处LBP特征描述并不有效。针对SAR图像相干斑噪声和图像局部梯度特性，项德良等[9]提出了基于局部梯度比率特征的二值模式。

 ${G_{{\rm{difference}}}}\left( {{g_p}} \right) = \left| {{g_p}-{g_{i, j}}} \right|$ (3)

 ${G_{{\rm{ratio}}}}\left( {{g_p}} \right) = \frac{{{G_{{\rm{difference}}}}\left( {{g_p}} \right)}}{{{g_p}}}$ (4)

 $\overline {{G_{{\rm{ratio}}}}\left( g \right)} = \frac{1}{P}\sum\limits_{p = 1}^P {{G_{{\rm{ratio}}}}\left( {{g_p}} \right)}$ (5)

 ${\rm{LGR}}{{\rm{P}}_{P, R}}\left( {{g_{i, j}}} \right) = \sum\limits_{P = 0}^{P-1} s \left( {{G_{{\rm{ratio}}}}\left( {{g_p}} \right)-\overline {{G_{{\rm{ratio}}}}\left( g \right)} } \right){2^P}$ (6)

 ${\rm{LGRPH}}\left( k \right) = \sum\limits_{i = 1}^N {\sum\limits_{j = 1}^M {f\left( {{\rm{LGR}}{{\rm{P}}_{P, R}}\left( {{g_{i, j}}} \right), k} \right)}, k \in \left[{0, K} \right]}$

 $f\left( {x, y} \right) = \left\{ \begin{array}{l} 1, \;\;\;\;x = y\\ 0, \;\;\;\;其他 \end{array} \right.$

 图 8 LGRP特征计算过程 Fig.8 Calculation process of the LGRP feature
2.2 旋转不变LGRP特征提取方法

1) 根据前文介绍的LGRP算子计算公式，计算得到所有邻域像素的GRP值，再由判别函数s(x)生成二进制编码串。

2) 对循环移位后可得到相同最小二进制模式的编码模式进行合并。以4位的二进制编码为例，编码1110(14)、1101(13)、1011(11) 和0111(7) 通过循环移位均可达到最小的编码0111。根据旋转不变LBP思路，这4种模式将会合并为一种新的模式。以8邻域采样点为例，合并过程如图 9所示。

 图 9 合并模式的映射关系 Fig.9 The mapping relation of merging pattern

3) 对于合并后得到的新模式，分别计算其值作为相应像素的特征值。

4) 计算得到旋转不变LGRP特征后，采用对称KL准则SKLD(symmetry kullback-leibler divergence)[11]来比较不同图像的特征。其流程图如图 10所示。

 图 10 相似度计算流程 Fig.10 Calculation process of the similarity
3 实验结果与分析 3.1 实验参数讨论

 图 11 参数P和Rmax对相似度曲线的影响 Fig.11 The influence of parameter P and Rmax on similarity curve

3.2 MSTAR目标识别实验

 图 12 同类目标比较和不同目标识别的相似度曲线 Fig.12 The similarity curves of similar objective comparison and different target recognition

3.3 SAR图像纹理识别实验

 图 13 四种不同角度的城区纹理切片 Fig.13 The texture slices of urban in different

 图 14 不同算子间性能比较相似度曲线 Fig.14 Performance comparison curve betweendifferent operators
4 结束语

 [1] SONG C, YANG F, LI P. Rotation invariant texture measured by local binary pattern for remote sensing image classification[C].Education Technology and Computer Science (ETCS), 2010 Second International Workshop on. IEEE, 2010, 3: 3-6. https://www.researchgate.net/publication/232641723_Rotation_Invariant_Texture_Measured_by_Local_Binary_Pattern_for_Remote_Sensing_Image_Classification (0) [2] OJALA T M, HARWOOD D. A comparative study of texture measures with classification based on featured distributions[J]. Pattern recognition, 1996, 29(1): 51-59. DOI:10.1016/0031-3203(95)00067-4 (0) [3] AHONEN T, HADID A M. Face description with local binary patterns: Application to face recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2006, 28(12): 2037-2041. DOI:10.1109/TPAMI.2006.244 (0) [4] AHONEN T, HADID A, M. Face recognition with local binary patterns[J]. European conference on computer vision, 2004, 3021(12): 469-481. (0) [5] NING J, ZHANG L, ZHANG D, et al. Robust object tracking using joint color-texture histogram[J]. International journal of pattern recognition and artificial intelligence, 2009, 23(07): 1245-1263. DOI:10.1142/S0218001409007624 (0) [6] Topi M, Timo O, Matti P, et al. Robust texture classification by subsets of local binary patterns[C]//Proceedings of the 15th International Conference on Pattern Recognition. 2000. IEEE, 2000, 3(3): 935-938. https://www.researchgate.net/publication/220931208_Robust_Texture_Classification_by_Subsets_of_Local_Binary_Patterns (0) [7] Ojala T, Pietikäinen M, Mäenpää T. Multiresolution gray-scale and rotation invariant texture classification with local binary patterns[J]. IEEE transactions on pattern analysis and machine intelligence, 2002, 24(7): 971-987. DOI:10.1109/TPAMI.2002.1017623 (0) [8] DAI D, YANG W, SUN H. Multilevel local pattern histogram for SAR image classification[J]. IEEE geoscience and remote sensing letters, 2011, 8(2): 225-229. DOI:10.1109/LGRS.2010.2058997 (0) [9] 项德良, 粟毅, 赵凌君, 等. 一种基于局部梯度比率特征度量SAR图像相似性的新方法[J]. 电子学报, 2014, 01: 9-13. XIANG Deliang, SU Yi, ZHAO Lingjun, et al. A new algorithm for SAR imagery similarity measure based on local gradient ratio pattern[J]. Acta electronica sinica, 2014, 01: 9-13. DOI:10.3969/j.issn.0372-2112.2014.01.002 (0) [10] GUAN Dongdong, TANG Tao, ZHAO Ling-jun, et al. A feature combining spatial and structural information for SAR image classification[C].Geoscience and Remote Sensing Symposium (IGARSS), 2015 IEEE International. IEEE, 2015: 4396-4399. http://ieeexplore.ieee.org/document/7326801/ (0) [11] LI L, TONG C S, CHOY S K. Texture classification using refined histogram[J]. IEEE transactions on image processing, 2010, 19(5). (0)