激光与光电子学进展, 2020, 57 (6): 061003, 网络出版: 2020-03-06
用于图匹配的子图学习算法 
下载： 1379次
A Subgraph Learning Method for Graph Matching
图 & 表
图 4. 离散实验的召回率和精确率。(a)离散值对召回率的影响;(b)离散值对精确率的影响
Fig. 4. Recall rate and precision in the outlier experiments. (a) Effect of discrete values on recall rate; (b) effect of discrete values on precision
图 5. 变形噪声实验的召回率和精度。(a)变形噪声对召回率的影响;(b)变形噪声对精确率的影响
Fig. 5. Recall rate and precision in the deformation noise experiments. (a) Effect of deformation noise on recall rate; (b) effect of deformation noise on precision
图 6. 不同边缘密度实验的召回率和精确率。(a)边缘密度对召回率的影响;(b)边缘密度对精确率的影响
Fig. 6. Recall rate and precision in the experiments with different edge densities. (a) Effect of edge density on recall rate; (b) effect of edge density on precision
图 7. Caltech+MSRC上的摩托车图形匹配样本。(a)基于SGM的匹配样本(正确匹配率为12/54);(b)基于RRWM的匹配样本(正确匹配率为11/67);(c)基于IPFP的匹配样本(正确匹配率为7/67);(d)基于SM的匹配样本(正确匹配率为9/67)
Fig. 7. Samples of graph matching for motorbike on Caltech+MSRC. (a) SGM-based matching sample (correct matching rate is 12/54); (b) RRWM-based matching sample (correct matching rate is 11/67); (c) IPFP-based matching sample (correct matching rate is 7/67); (d) SM-based matching sample (correct matching rate is 9/67)
图 8. Caltech+MSRC帽子图形匹配样本。(a)基于SGM的匹配样本(正确匹配率为4/7);(b)基于RRWM的匹配样本(正确匹配率为4/9);(c)基于IPFP的匹配样本(正确匹配率为2/9);(d)基于SM的匹配样本(正确匹配率为2/9)
Fig. 8. Samples of graph matching for cap on Caltech+MSRC. (a) SGM-based matching sample (correct matching rate is 4/7); (b) RRWM-based matching sample (correct matching rate is 4/9); (c) IPFP-based matching sample (correct matching rate is 2/9); (d) SM-based matching sample (correct matching rate is 2/9)
图 9. Caltech+MSRC汽车图形匹配样本。(a)基于SGM的匹配样本(正确匹配率为16/30);(b)基于RRWM的匹配样本(正确匹配率为12/36);(c)基于IPFP的匹配样本(正确匹配率为4/36);(d)基于SM的匹配样本(正确匹配率为4/36)
Fig. 9. Samples of graph matching for car on Caltech+MSRC. (a) SGM-based matching sample (correct matching rate is 16/30); (b) RRWM-based matching sample (correct matching rate is 12/36); (c) IPFP-based matching sample (correct matching rate is 4/36); (d) SM-based matching sample (correct matching rate is 4/36)
表 1不同方法在Caltech+MSRC上的召回率和精确率
Table1. Recall rate and precision of different methods on Caltech+MSRC
|
陈闯, 王亚, 贾文武. 用于图匹配的子图学习算法[J]. 激光与光电子学进展, 2020, 57(6): 061003. Chuang Chen, Ya Wang, Wenwu Jia. A Subgraph Learning Method for Graph Matching[J]. Laser & Optoelectronics Progress, 2020, 57(6): 061003.
PDF全文
