改进的半监督协同SOM图匹配算法
|
摘要
为进一步解决目前非精确图匹配算法在应用上的局限性,提出一种基于半监督的协同自组织映射神经网络(selforganizing map)算法。通过挖掘图的完备特征空间集,将有限带标签样本均分为3个训练集,分别初始化SOM子分类器,训练结果采用投票法进行统计;根据统计结果把无争议样本添加到已知标记的训练集中,票数不同的样本通过K均值算法进行二次判别。即在半监督条件下,利用聚类划分的思想将图匹配问题转化为图的分类标记问题。实验结果表明,该算法能够有效提升准确率,避免了传统算法在训练过程中产生的无法收敛的问题,节约了有监督算法的人工标记成本。
To solve the non-exact graph matching algorithm in the application of the limitations,a semi-supervised self-organizing map algorithm was proposed.By digging the complete set of feature space,the limited number of labeled samples was divided into three sub-training sets,and these labeled samples were used to initialize the SOM sub classifier.The voting method was used to train results for statistics analysis.Based on the statistical results,the uncontroversial samples were added to the training set of known markers,and the samples with different numbers of votes were judged using the K-means algorithm for the second time.Under the semi-supervised condition,the problem of classification and matching was transformed into the problem of classification and marking based on the idea of clustering.Through the example verification,this algorithm can effectively improve the accuracy of matching and avoid the non-convergence problem caused by the traditional algorithm in the training process.At the same time,it also solves the cost of artificial labeling with supervised algorithm.
To solve the non-exact graph matching algorithm in the application of the limitations,a semi-supervised self-organizing map algorithm was proposed.By digging the complete set of feature space,the limited number of labeled samples was divided into three sub-training sets,and these labeled samples were used to initialize the SOM sub classifier.The voting method was used to train results for statistics analysis.Based on the statistical results,the uncontroversial samples were added to the training set of known markers,and the samples with different numbers of votes were judged using the K-means algorithm for the second time.Under the semi-supervised condition,the problem of classification and matching was transformed into the problem of classification and marking based on the idea of clustering.Through the example verification,this algorithm can effectively improve the accuracy of matching and avoid the non-convergence problem caused by the traditional algorithm in the training process.At the same time,it also solves the cost of artificial labeling with supervised algorithm.
引文
[1]Carletti V,Foggia P,Vento M.Performance comparison of five exact graph matching algorithms on biological databases[G].LNCS 8158:International Conference on Image Analysis&Processing.Berlin:Springer Berlin Heidelberg,2013:409-417.
[2]Pinheiro MA,Kybic J,Fua P.Geometric graph matching using monte Carlo tree search[J].IEEE Trans Pattern Anal Mach Intell,2017,11(39):2171-2185.
[3]Luo Yong,Chen Shuwei,He Xiaojuan,et al.Alphanumeric character recognition based on BP neural network classification and combined features[J].International Journal of Computational Intelligence Systems,2013,6(6):1108-1115.
[4]Bougleux S,Brun L,Carletti V,et al.Graph edit distance as aquadratic assignment problem[J].Pattern Recognition Letters,2017,87(C):38-46.
[5]Yang J,Chu D,Zhang L,et al.Sparse representation classifier steered discriminative projection with applications to face recognition[J].IEEE Transactions on Neural Networks&Learning Systems,2013,24(7):1023-1035.
[6]Chiaradia,Alain J.Intelligibility in space syntax maze[J].Respiratory Research,2013,7(4):1-10.
[7]Maung C,Schweitzer H.Improved greedy algorithms for sparse approximation of a matrix in terms of another matrix[J].IEEE Transactions on Knowledge&Data Engineering,2015,27(3):769-780.
[8]Chauhan SK,Pradhan B.Self-organizing clustering methods for energy-efficient data gathering in sensor networks[J].International Journal of Engineering Trends&Technology,2014,14(5):68-75.
[9]Wang LG,Yang YS,Lu TT.Semi-supervised classification for hyperspectral image based on tri-training[J].Applied Mechanics&Materials,2014(687-691):3644-3647.
[10]Yang S,Yu J,Liu Y.A research of data stratification algorithm based on semi-supervised clustering[C]//IEEE International Conference on Progress in Informatics and Computing,2016:196-200.
[11]Serratosa F.Fast computation of Bipartite graph matching[J].Pattern Recognition Letters,2014,45(8):244-250.
[12]Wang X,Zhang C,Bai Y.Two modified sparse subspace clustering[J].International Journal of Mathematical and Computational Methods,2016(1):400-404.
[2]Pinheiro MA,Kybic J,Fua P.Geometric graph matching using monte Carlo tree search[J].IEEE Trans Pattern Anal Mach Intell,2017,11(39):2171-2185.
[3]Luo Yong,Chen Shuwei,He Xiaojuan,et al.Alphanumeric character recognition based on BP neural network classification and combined features[J].International Journal of Computational Intelligence Systems,2013,6(6):1108-1115.
[4]Bougleux S,Brun L,Carletti V,et al.Graph edit distance as aquadratic assignment problem[J].Pattern Recognition Letters,2017,87(C):38-46.
[5]Yang J,Chu D,Zhang L,et al.Sparse representation classifier steered discriminative projection with applications to face recognition[J].IEEE Transactions on Neural Networks&Learning Systems,2013,24(7):1023-1035.
[6]Chiaradia,Alain J.Intelligibility in space syntax maze[J].Respiratory Research,2013,7(4):1-10.
[7]Maung C,Schweitzer H.Improved greedy algorithms for sparse approximation of a matrix in terms of another matrix[J].IEEE Transactions on Knowledge&Data Engineering,2015,27(3):769-780.
[8]Chauhan SK,Pradhan B.Self-organizing clustering methods for energy-efficient data gathering in sensor networks[J].International Journal of Engineering Trends&Technology,2014,14(5):68-75.
[9]Wang LG,Yang YS,Lu TT.Semi-supervised classification for hyperspectral image based on tri-training[J].Applied Mechanics&Materials,2014(687-691):3644-3647.
[10]Yang S,Yu J,Liu Y.A research of data stratification algorithm based on semi-supervised clustering[C]//IEEE International Conference on Progress in Informatics and Computing,2016:196-200.
[11]Serratosa F.Fast computation of Bipartite graph matching[J].Pattern Recognition Letters,2014,45(8):244-250.
[12]Wang X,Zhang C,Bai Y.Two modified sparse subspace clustering[J].International Journal of Mathematical and Computational Methods,2016(1):400-404.