|
发布时间: 2019-05-16 |
图像理解和计算机视觉 |
|
|
收稿日期: 2018-06-22; 修回日期: 2018-10-31
基金项目: 国家自然科学基金项目(61101057)
第一作者简介:
孟琭, 1982年生, 男, 工学博士, 副教授, 主要研究方向为人工智能、图像处理、区块链。E-mail:menglu1982@gmail.com;
魏子然, 男, 本科生, 主要研究方向为图像处理。E-mail:523865970@qq.com.
中图法分类号: TP301.6
文献标识码: A
文章编号: 1006-8961(2019)05-0794-11
|
摘要
目的 现有的图匹配算法大多应用于二维图像,对三维图像的特征点匹配存在匹配准确率低和计算速度慢等问题。为解决这些问题,本文将分解图匹配算法扩展应用在了三维图像上。方法 首先将需要匹配的两个三维图像的特征点作为图的节点集;再通过Delaunay三角剖分算法,将三维特征点相连,则相连得到的边就作为图的边集,从而建立有向图;然后,根据三维图像的特征点构建相应的三维有向图及其邻接矩阵;再根据有向图中的节点特征和边特征分别构建节点特征相似矩阵和边特征相似矩阵;最后根据这两个特征矩阵将节点匹配问题转化为求极值问题并求解。结果 实验表明,在手工选取特征点的情况下,本文算法对相同三维图像的特征点匹配有97.56%的平均准确率;对不同三维图像特征点匹配有76.39%的平均准确率;在三维图像有旋转的情况下,有90%以上的平均准确率;在特征点部分缺失的情况下,平均匹配准确率也能达到80%。在通过三维尺度不变特征变换(SIFT)算法得到特征点的情况下,本文算法对9个三维模型的特征点的平均匹配准确率为98.78%。结论 本文提出的基于图论的三维图像特征点匹配算法,经实验结果验证,可以取得较好的匹配效果。
关键词
图匹配; 三维图像处理; 图论; 路径跟踪算法; 人工智能
Abstract
Objective Graph matching involves establishing a one-to-one correspondence between the feature points of two images on the basis of graph theory. As a basic problem in computer vision, graph matching is closely related to many computer vision areas, such as object tracking, image classification, object recognition, contour matching, and so on. Existing graph matching algorithms are mostly applied to two-dimensional images, and many problems, such as low accuracy rate and slow calculation speed, remain in matching feature points from three-dimensional images. To solve these problems, this study generalized the factorized graph matching algorithm to three-dimensional images. Method The three-dimensional graph matching method comprises five main steps. In the first step, the feature points of the two three-dimensional images that must be matched are used as the node set of the graph. Then, the three-dimensional feature points are connected by the Delaunay triangulation algorithm, and the obtained edges are used as the edge sets of the graph to establish the directed graph. In the second step, the starting node matrix is computed on the basis of the structure of the directed graph, and each element with a value of 1 represents the starting node of the edge in the graph. The ending node matrix can be computed in the same manner. The starting node matrix and the ending node matrix can be combined to represent which node of each edge in the graph is initiated and which node is terminated. In the third step, we use the graph's degree and eccentricity as the node's feature to build node feature vectors. We also use the edge length and the angle between the edge and the plane XOY as the edge's feature to build edge feature vectors. The node feature adjacency matrix can be calculated according to the node feature vectors. The number of matrix rows is equal to the number of one graph's nodes. The number of matrix columns is equal to the number of the other graph's nodes. The value of each element in the matrix can be calculated on the basis of the Euclidean distance between corresponding nodes from the two graphs and represents the degree of similarity between two nodes in the graphs. Similarly, the edge feature adjacency matrix can be calculated on the basis of edge feature vectors. The node feature adjacency matrix and edge feature adjacency matrix can be used to delineate the degree of similarity between nodes and edges from two graphs. In the fourth step, using the starting node matrix, ending node matrix, node feature adjacency matrix, and edge feature adjacency matrix, we can build a special matrix K to transform the three-dimensional graph matching problem into a problem of solving a node matching matrix X to maximize an equation J. In the fifth step, the maximum value solution of equation J is a quadratic assignment Problem; therefore, we use the path following algorithm to obtain the optimal solution by splitting the equation J into a series of concave and convex expressions and then find the optimal solution X by iterative approximation. The solution matrix X is the node matching matrix and represents the feature point matching result from two three-dimensional images. Result We tested and validated our method in five experiments. In the first experiment, we manually picked 102 feature points from a three-dimensional image in the dataset that comprised nine three-dimensional car images and had a 97.56% average accuracy rate in feature point matching from the same three-dimensional images. In the second experiment, we achieved a 76.39% average accuracy rate from the feature point matching from different three-dimensional images from the dataset. In the third experiment, we rotated the three-dimensional images every 10° from 10° to 180° and matched all 102 manually selected feature points from the same three-dimensional images; the average accuracy rate of matching was 90%. In the fourth experiment, we randomly removed several feature points from the three-dimensional images. During the removal, the accuracy rate of matching decreased. Nevertheless, removing a maximum of 30 feature points still yielded an 80% accuracy rate of matching. In the first four experiments, we used the manually selected feature points from the three-dimensional images; furthermore, we validated our method on the feature points automatically obtained by the 3D scale-invariant feature transform (SIFT) algorithm and achieved a 98.78% average accuracy rate in the feature point matching of the three-dimensional images. Conclusion The three-dimensional image feature point matching algorithm based on graph theory proposed in this work was tested and validated by five experiments. Results indicate that our method can achieve good matching results for manually selected feature points and automatically calculated feature points.
Key words
graph matching; three-dimensional image processing; graph theory; path following algorithm; artificial intelligence
0 引言
图像特征点是用来表示图像中具有特殊含义或代表性的节点,如人体的关节点、物体梯度的变化点、目标的感兴趣点等。图像特征点匹配可以对两个图像中的特征点建立一一对应关系,从而为识别图像中物体的位置、姿态、甚至动作打下基础,也使其在对象跟踪、全息影像创作、图像分类、图像对齐等领域处于关键位置[1-2]。
图论作为数学的一个分支,在20世纪70年代被提出应用到图像匹配问题中,由于需要巨大的运算量,而当时的客观条件无法满足需求,在经历了最初对这个巧妙理论结构的热情后,这方面的研究被搁置了很长一段时间。随着计算机技术的不断发展,近几年,图论理论在图像匹配中的应用再次被提了出来,并且有了不小的进展。
2005年,Berg等人[3]提出了二维图像匹配算法,将图像匹配问题设定为一个整体二次规划问题。其中的价值函数是基于对应几何模糊点描述符的相似性,例如相应特征点的几何失真等。该算法使基于图论的图像匹配得以实现,并且对异常值也有一定的掌控能力。2008年,Mateus等人[4]提出了一种解决三维图像匹配的方法,使用权重图描述由像素组成的清晰的形状,图像匹配就可以被归纳为最大子图同构问题。将稠密点通过无监督聚类和EM算法,在正交变换下转换为图,通过用直方图建立的特征函数为图附上权重。在此基础上,通过选择拉普拉斯矩阵的特征函数的最佳子集来寻找两个相对应的
综上所述,目前的图匹配算法大多面向二维图像,面向三维图像时,缺少既准确又快速的匹配算法。本文拟将分解图匹配算法(FGM)从现在仅适用于二维图像,推广到适用于三维图像。在众多图匹配算法中选择FGM,是因为在二维图像特征点匹配过程中,FGM表现出了更好的准确率和更快的速度。
1 方法
1.1 算法总体流程
面向三维图像的图匹配算法总体流程图如图 1所示,首先将需要匹配的两个三维图像的特征点作为图的节点集(vertices set)。然后通过Delaunay三角剖分算法,将三维特征点相连,得到的边作为图的边集(edges set),从而建立有向图。令
1.2 建立有向图及起点、终点特征矩阵
令需要进行特征点匹配的两个三维图像分别为
令
$ {\mathit{\boldsymbol{G}}_1} = \left[ {\begin{array}{*{20}{l}} 1&0&0&0&0&0\\ 0&1&0&1&0&0\\ 0&0&1&0&1&0\\ 0&0&0&0&0&1 \end{array}} \right] $ |
$ {\mathit{\boldsymbol{H}}_1} = \left[ {\begin{array}{*{20}{c}} 0&0&0&1&0&0\\ 1&0&0&0&1&0\\ 0&1&0&0&0&1\\ 0&0&1&0&0&0 \end{array}} \right] $ |
$ {\mathit{\boldsymbol{G}}_2} = \left[ {\begin{array}{*{20}{c}} 1&0&0&0\\ 0&1&1&0\\ 0&0&0&1 \end{array}} \right], \;\;\, {\mathit{\boldsymbol{H}}_2} = \left[ {\begin{array}{*{20}{c}} 0&0&1&0\\ 1&0&0&1\\ 0&1&0&0 \end{array}} \right] $ | (1) |
1.3 建立特征相似矩阵
令
$ \left\{ {\begin{array}{*{20}{l}} {{\mathit{\boldsymbol{P}}_1} = \left[ {p_1^1, \cdots , p_{{n_1}}^1} \right] \in {{\bf{R}}^{{d_p} \times {n_1}}}}\\ {{\mathit{\boldsymbol{P}}_2} = \left[ {p_1^2, \cdots , p_{{n_2}}^2} \right] \in {{\bf{R}}^{{d_p} \times {n_2}}}}\\ {{\mathit{\boldsymbol{Q}}_1} = \left[ {q_1^1, \cdots , q_{{m_1}}^1} \right] \in {{\bf{R}}^{{d_q} \times {m_1}}}}\\ {{\mathit{\boldsymbol{Q}}_2} = \left[ {q_1^1, \cdots , q_{{m_2}}^1} \right] \in {{\bf{R}}^{{d_q} \times {m_2}}}} \end{array}} \right. $ | (2) |
式中,
本文的节点特征选取了度和离心率来表示[10]。度是指某一节点连接的边的数量;离心率
$ {\varepsilon _i} = \mathop {\max }\limits_{{n_j} \in \mathit{\boldsymbol{N}}} \left( {{H_{{n_i} \to {n_j}}}} \right) $ | (3) |
$ H_{A \rightarrow B}=\min\limits _{k \in\{1, \cdots, N-1\}}\left(P_{A \rightarrow B}(k)\right) $ | (4) |
本文的边节点特征选取了边的长度
$ l=\sqrt{\left(x_{1}-x_{2}\right)^{2}+\left(y_{1}-y_{2}\right)^{2}+\left(z_{1}-z_{2}\right)^{2}} $ | (5) |
$ \begin{array}{c}{\theta=} \\ {\text { arcos } \frac{\sqrt{\left(x_{1}-x_{2}\right)^{2}+\left(y_{1}-y_{2}\right)^{2}}}{\sqrt{\left(x_{1}-x_{2}\right)^{2}+\left(y_{1}-y_{2}\right)^{2}+\left(z_{1}-z_{2}\right)^{2}}}}\end{array} $ | (6) |
式中,
通过
1.4 计算节点对应关系矩阵
有了起点特征矩阵
$ {{J}_{\text{gm}}}(X)=\sum\limits_{{{i}_{1}}{{i}_{2}}}{{{x}_{{{i}_{1}}{{i}_{2}}}}}k_{{{i}_{1}}{{i}_{2}}}^{p}+\sum\limits_{\begin{smallmatrix} {{i}_{1}}\ne {{i}_{2}},{{j}_{1}}\ne {{j}_{2}} \\ \text{g}_{{{i}_{1}}{{c}_{1}}}^{1}h_{{{j}_{1}}{{c}_{1}}}^{1}=1n \\ \text{g}_{{{i}_{2}}{{c}_{2}}}^{2}h_{{{j}_{2}}{{c}_{2}}}^{2}=1 \end{smallmatrix}}{{{x}_{{{i}_{1}}{{i}_{2}}}}}{{x}_{{{j}_{1}}{{j}_{2}}}}k_{{{c}_{1}}{{c}_{2}}}^{q} $ | (7) |
的最大值。若想要式(7)取最大值,则其中的两个部分都需要得到最大值。第1部分中的
为了求解式(7),定义一个相似关系矩阵
$ \begin{array}{l} \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{k_{{i_1}{i_2}, {j_1}{j_2}}} = \\ \left\{ {\begin{array}{*{20}{l}} {k_{{i_1}{i_2}}^p\;\;\;若{i_1} = {j_1}且{i_2} = {j_2}}\\ {k_{{c_1}{c_2}}^q\;\;\;若{i_1} \ne {j_1}且{i_2} \ne {j_2}且g_{{i_1}{c_1}}^1h_{{j_1}{c_1}}^1g_{{i_2}{c_2}}^2h_{{j_2}{c_2}}^2 = 1}\\ {0\;\;\;\;\;其他} \end{array}} \right.\\ \end{array} $ | (8) |
矩阵
有了矩阵
$ \mathop {\max }\limits_X {J_{{\rm{gm}}}}(X) = {\mathop{\rm vec}\nolimits} {(\mathit{\boldsymbol{X}})^{\rm{T}}}\mathit{\boldsymbol{K}}{\mathop{\rm vec}\nolimits} (\mathit{\boldsymbol{X}}) $ | (9) |
式中,
$ \begin{array}{*{20}{c}} {\mathit{\boldsymbol{K}} = {\mathop{\rm diag}\nolimits} \left( {{\mathop{\rm vec}\nolimits} \left( {{\mathit{\boldsymbol{K}}_p}} \right)} \right) + }\\ {\left( {{\mathit{\boldsymbol{G}}_1} \otimes {\mathit{\boldsymbol{G}}_2}} \right){\mathop{\rm diag}\nolimits} \left( {{\mathop{\rm vec}\nolimits} \left( {{\mathit{\boldsymbol{K}}_q}} \right)} \right){{\left( {{\mathit{\boldsymbol{H}}_2} \otimes {\mathit{\boldsymbol{H}}_1}} \right)}^{\rm{T}}}} \end{array} $ | (10) |
式中,
$ \begin{array}{*{20}{c}} {\mathop {\max }\limits_X {J_{{\rm{gm}}}}(\mathit{\boldsymbol{X}}) = }\\ {{\mathop{\rm tr}\nolimits} \left( {\mathit{\boldsymbol{K}}_p^{\rm{T}}\mathit{\boldsymbol{X}}} \right) + {\mathop{\rm tr}\nolimits} \left( {\mathit{\boldsymbol{K}}_q^{\rm{T}}\left( {\mathit{\boldsymbol{G}}_1^{\rm{T}}\mathit{\boldsymbol{X}}{\mathit{\boldsymbol{G}}_2^ \circ} \mathit{\boldsymbol{H}}_1^{\rm{T}}\mathit{\boldsymbol{X}}{\mathit{\boldsymbol{H}}_2}} \right)} \right)} \end{array} $ | (11) |
式中,
$ \mathop {\max }\limits_X {J_{{\rm{gm}}}}(\mathit{\boldsymbol{X}}) = {\mathop{\rm tr}\nolimits} \left( {\mathit{\boldsymbol{K}}_p^{\rm{T}}\mathit{\boldsymbol{X}}} \right) + \sum\limits_{i = 1}^c {{\mathop{\rm tr}\nolimits} } \left( {\mathit{\boldsymbol{A}}_i^1\mathit{\boldsymbol{XA}}_i^2{\mathit{\boldsymbol{X}}^{\rm{T}}}} \right) $ | (12) |
式中,
式(12)的求解是一个二次分配问题(QAP),这类问题的求解一直是数学上的难题,本文采用路径跟踪算法(PFA)[11]求其近似解。路径跟踪算法是通过将待求目标转化为一系列凹、凸表达的累加,并通过迭代逼近的方式求得最优解。因此,本文需将式(12)转化为凹、凸表达,为此引入常数
$ \gamma = \sum\limits_{i = 1}^c {{\mathop{\rm tr}\nolimits} } \left( {\mathit{\boldsymbol{A}}_i^{1{\rm{T}}}\mathit{\boldsymbol{X}}{\mathit{\boldsymbol{X}}^{\rm{T}}}\mathit{\boldsymbol{A}}_i^1} \right) + {\mathop{\rm tr}\nolimits} \left( {\mathit{\boldsymbol{A}}_i^2\mathit{\boldsymbol{X}}{\mathit{\boldsymbol{X}}^{\rm{T}}}\mathit{\boldsymbol{A}}_\mathit{i}^{2{\rm{T}}}} \right) $ | (13) |
则有
$ {J_{{\rm{vex}}}}(\mathit{\boldsymbol{X}}) = {J_{{\rm{gm}}}}(\mathit{\boldsymbol{X}}) - \frac{\gamma }{2} $ |
$ {J_{{\rm{cav}}}}(\mathit{\boldsymbol{X}}) = {J_{{\rm{gm}}}}(\mathit{\boldsymbol{X}}) - \frac{\gamma }{2} $ |
式中,
$ \begin{array}{l} \mathop {\max }\limits_X {J_\alpha }(\mathit{\boldsymbol{X}}) = (1 - \alpha ){J_{{\rm{vex}}}}(\mathit{\boldsymbol{X}}) + \alpha {J_{{\mathop{\rm cav}\nolimits} }}(\mathit{\boldsymbol{X}}) = \\ \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{J_{{\rm{gm}}}}(\mathit{\boldsymbol{X}}) + \left( {\alpha - \frac{1}{2}} \right)\gamma \end{array} $ | (14) |
式中,
1) 将
2) 将
3) 应用Frank-Wolfe’s算法[12],计算得到最大值
4) 若
5) 重复步骤2), 直到函数收敛或达到指定迭代次数为止。
至此,求解得到的矩阵
2 实验
2.1 实验数据
实验使用的三维图像数据是从网上(https://free3d.com)免费下载,共下载了9个汽车的三维图像模型,并导入三维空间坐标系中,如图 3所示。随后在这些三维图像上标记特征点,本实验手动标记了102个特征点,特征点集中不包含错误特征点,在每个轮子上标记了4个特征点;在车的前侧与后侧分别标记了4行特征点,每行3个点;在左右增加了一些点来更细致地描述轮廓,如图 4所示。
2.2 实验结果
本实验共分为5个部分,前4个是对手工选取的特征点进行匹配,第5个是对算法自动生成的特征点进行匹配,具体为:1)对相同的图像进行特征点匹配;2)对不同图像进行特征点匹配;3)对相同的图像旋转不同角度后,再进行特征点匹配;4)对相同的图像随机去掉一定数量的节点后,再进行特征点匹配;5)对3D SIFT算法得到的特征点进行匹配。
2.2.1 对相同三维图像进行特征点匹配
2.2.2 对不同三维图像进行特征点匹配
对三维图像数据集中的每两个图像之间都进行特征点匹配,因为数据集都是各种车型的三维图像,各种车型之间存在一定的相似性,因此可以进行匹配,此时的匹配依据仍然是三维图起点、终点特征矩阵以及三维图特征相似矩阵,同时又考虑到不同车型的三维图像终归会存在一定程度上的差异性,这会使得一部分本该匹配在一起的特征点因为局部结构的差异性,从而引起一定程度的误匹配。两两之间的匹配结果如表 2所示,平均准确率为76.39%。值得注意的是,表 2中的数据呈现对称的模式,因为从第
表 2
不同三维图像特征点匹配结果
Table 2
Matching results of feature points in different 3D images
/% | |||||||||
序号 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |
1 | 62 | 77 | 69 | 77 | 75 | 77 | 87 | 90 | |
2 | 62 | 75 | 59 | 64 | 78 | 93 | 74 | 72 | |
3 | 77 | 75 | 73 | 68 | 80 | 78 | 86 | 78 | |
4 | 69 | 59 | 73 | 68 | 63 | 61 | 78 | 67 | |
5 | 77 | 64 | 68 | 68 | 66 | 83 | 76 | 86 | |
6 | 75 | 78 | 80 | 63 | 66 | 68 | 93 | 84 | |
7 | 77 | 93 | 78 | 61 | 83 | 68 | 76 | 93 | |
8 | 87 | 74 | 86 | 78 | 76 | 93 | 76 | 96 | |
9 | 90 | 72 | 78 | 67 | 86 | 84 | 93 | 96 |
2.2.3 对相同三维图像旋转不同角度进行特征点匹配
为了测试本文算法对三维图像刚性变化情况下的匹配准确率,将三维图像绕
2.2.4 对相同三维图像随机去掉若干特征点进行特征点匹配
为了测试本文算法对三维图像非刚性变化情况下的匹配准确率,随机去掉1~30个特征点后再进行匹配。由于每次匹配时,被匹配的三维图像中去掉的特征点是随机的,所以不是所有的特征点都是一一对应的,因此匹配准确率的计算与前3个实验略有不同,由匹配正确的特征点对的数量与可以匹配的特征点对的数量的比值作为准确率。随机去掉若干特征点后,再进行特征点匹配的准确率曲线如图 9所示,可以看到,随着随机去掉特征点的数目的增加,匹配的准确率也在逐渐下降,这是因为待匹配的特征点受到随机删除的影响,使得本该与之最佳匹配的点变成了非最佳,从而导致误匹配。但是即使是去掉了30个特征点的情况下,匹配准确率也在70%以上。说明本文算法受三维图像非刚性变化的影响较小。其中一个三维图像数据在随机去掉10个、20个、30个特征点后的匹配结果如图 10所示,其中绿色线连接的是匹配正确的节点对,红色线连接的是匹配错误的节点对。
2.2.5 对三维SIFT算法得到的特征点进行特征点匹配
之前的验证过程中,特征点都是手工标注的,为了进一步验证本文算法,通过三维SIFT算法[13]计算得到36个特征点,并应用本文算法对这些三维特征点进行匹配。选取的三维图像仍然是前文所述的9个汽车模型,图 11展示了其中一个的特征点匹配结果,准确率为100%。所有9个三维汽车模型的特征点匹配准确率结果如表 3所示,平均匹配准确率为98.78%。可以验证,本文算法对手工标注的三维特征点和特征提取算法得到的三维特征点的匹配效果都很好。
表 3
9个三维汽车模型的SIFT特征点匹配准确率
Table 3
Accuracy of SIFT feature point matching for nine 3D vehicle models
序号 | |||||||||
1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | |
准确率/% | 100 | 98 | 98 | 99 | 98 | 99 | 100 | 98 | 99 |
3 结论
本文面向三维图像数据,提出了三维图匹配算法,解决了现有图匹配算法仅适用于二维图像特征点匹配的问题。实验表明,本文算法对相同三维图像的特征点匹配有97.56%的平均准确率;对不同三维图像特征点匹配有76.11%的平均准确率;在三维图像有旋转的情况下,有90%以上的平均准确率;在特征点部分缺失的情况下,平均匹配准确率也能达到80%。综上,本文提出的三维图匹配算法可以很好地对三维图像的特征点进行匹配,但是在面向不同三维图像特征点匹配的时候,其准确率还有待提高。今后的研究方向是对刚性及非刚性变换条件下的三维图像做特征点匹配。
参考文献
-
[1] Jiang B, Tang J, Luo B. A survey on graph matching algorithms in computer vision[J]. Journal of Anhui University:Natural Science Edition, 2017, 41(1): 29–36. [江波, 汤进, 罗斌. 计算机视觉中的图匹配方法研究综述[J]. 安徽大学学报:自然科学版, 2017, 41(1): 29–36. ] [DOI:10.3969/j.issn.1000-2162.2017.01.005]
-
[2] Xu J, Zhang Q Z, Zhao X, et al. Survey on dynamic graph pattern matching technologies[J]. Journal of Software, 2018, 29(3): 663–688. [许嘉, 张千桢, 赵翔, 等. 动态图模式匹配技术综述[J]. 软件学报, 2018, 29(3): 663–688. ] [DOI:10.13328/j.cnki.jos.005444]
-
[3] Berg A C, Berg T L, Malik J. Shape matching and object recognition using low distortion correspondences[C]//Proceedings of 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. San Diego, CA, USA: IEEE, 2005: 26-33. [DOI:10.1109/CVPR.2005.320]
-
[4] Mateus D, Horaud R, Knossow D, et al. Articulated shape matching using laplacian eigenfunctions and unsupervised point registration[C]//Proceedings of 2008 IEEE Conference on Computer Vision and Pattern Recognition. Anchorage, AK, USA: IEEE, 2008: 1-8. [DOI:10.1109/CVPR.2008.4587538]
-
[5] Jiang H, Yu S X, Martin D R. Linear scale and rotation invariant matching[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2011, 33(7): 1339–1355. [DOI:10.1109/TPAMI.2010.212]
-
[6] Han F. 3D ear shape feature matching based on graph theory[D]. Dalian: Liaoning Normal University, 2014. [韩枫.基于图论的三维耳廓形状特征匹配[D].大连: 辽宁师范大学, 2014.]
-
[7] Zhou F, De la Torre F. Factorized graph matching[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(9): 1774–1789. [DOI:10.1109/TPAMI.2015.2501802]
-
[8] Wang T, Ling H B, Lang C Y, et al. Symmetry-aware graph matching[J]. Pattern Recognition, 2016, 60: 657–668. [DOI:10.1016/j.patcog.2016.06.022]
-
[9] Dou H, Baker M L, Ju T. Graph-based deformable matching of 3D line with application in protein fitting[J]. The Visual Computer, 2015, 31(6-8): 967–977. [DOI:10.1007/s00371-015-1115-x]
-
[10] Balakrishnan R, Ranganathan K. A Textbook of Graph Theory[M]. New York: Springer, 2011.
-
[11] Zaslavskiy M, Bach F, Vert J P. A path following algorithm for the graph matching problem[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2009, 31(12): 2227–2242. [DOI:10.1109/TPAMI.2008.245]
-
[12] Cai H, He R C, Ma C X, et al. Improved Frank-Wolfe algorithm for path traffic flows in traffic assignment problems[J]. Computer Engineering and Applications, 2018, 54(9): 213–217. [柴获, 何瑞春, 马昌喜, 等. 用于求解路径交通流量的改进Frank-Wolfe算法[J]. 计算机工程与应用, 2018, 54(9): 213–217. ] [DOI:10.3778/j.issn.1002-8331.1612-0102]
-
[13] Darom T, Keller Y. Scale-invariant features for 3D mesh models[J]. IEEE Transactions on Image Processing, 2012, 12(5): 2758–2769. [DOI:10.1109/TIP.2012.2183142]