连续蒙特卡洛算法下图像匹配问题的研究.pdfVIP

2017 年5 月 控 制 工 程 May. 2017 第24 卷第5 期 Control Engineering of China Vol.24, No.5 文章编号：1671-7848(2017)05-0973-05 DOI: 10.14107/j.cnki.kzgc.150362 连续蒙特卡洛算法下图像匹配问题的研究 刘美菊，杨宏钰，李凌燕，高恩阳 (沈阳建筑大学 信息与控制工程学院，沈阳 110168) 摘 要：针对传统形状上下文算法在图像内部变形，发生遮挡的情况下匹配率较低的问题， 提出一种基于形状上下文的连续蒙特卡洛算法。对其准确性和时效性进行验证。并和算法 Sift，PA ，ULA,TBA 进行对比。首先，对图像特征点分别计算点模式形状上下文信息。利 用卡方距离得到初匹配。其次，利用初匹配结果构建图模型，构造亲近矩阵，将亲近矩阵 进行整数二次规划，以得出最优匹配方案。最后，对最优匹配方案进行图模型的重构与连 续分配。实验结果表明，所提出的算法在保证较高图像匹配率的同时，对图像视角不同和 尺度不同以及内部遮挡的条件具有很好的鲁棒性。 关键词：图像匹配；连续蒙特卡洛算法；形状上下文；卡方距离；亲近矩阵 中图分类号：TP39 文献标识码：A An Algorithm for Graph Matching Based on Sequential Monte Carlo LIU Mei-ju, YANG Hong-yu, LI Ling-yan, GAO En-yang (College of Information and Control Engineering, Shenyang Jianzhu University, Shenyang 110168, China) Abstract: Aiming at the problem that the graph matching ratio is very low in traditional shape context algorithm in circumstances of deformation and occlusion, a sequential Monte Carlo algorithm is proposed. The accuracy and timeliness of the algorithm is verified and compared with the algorithm Sift, PA, ULA and TBA. Firstly, the shape context information of each point is calculated respectively, and the Chi-Square distance is used to acquire the initial matching. Secondly, the initial matching is used to construct the graph model, and the affinity matrix is constructed based on the graph model, the affinity matrix is brought into the integer quadratic program in order to acquire the best matching program. Finally, the graph model is reconstructed and