基于迭代重加权的高阶张量图匹配算法

摘要

Graph matching is a fundamental and significant problem in computer vision.Sparse constraint is widely used in machine learning and image processing as an effective optimization method.The classical graph matching method can't obtain sufficient effective and sparse approximate solution.In order to solve this problem and further explore the application of sparse optimization in graph matching,a L1/2 norm method is introduced to improve the high order tensor graph matching model.And the proposed method is exploited to approximate the non-convex non-smooth model based on iterative weighting.The comparison experiment on the standard benchmark data set shows that the higher order graph matching algorithm based on iterative weighting can obtain more efficient and sparsely strong solution and improve the matching accuracy.Meanwhile,it is more robust on the promising performance of resisting matching noise point than conventional algorithms.%图匹配是计算机视觉中基础且重要的一个问题.稀疏约束作为一种有效的优化方法,被广泛应用于机器学习和图像处理中.传统的图匹配方法并不能获得足够有效且稀疏的近似解,为解决这个问题且进一步探究稀疏优化在图匹配中的应用,故引入一种L1/2范数以改进高阶张量图匹配模型,并提出了基于迭代重加权的方法以近似求解该非凸非光滑模型.通过标准实验数据集上的对比实验表明,基于迭代重加权的高阶图匹配算法可以得到更加有效且稀疏性强的解,提高了匹配准确率.同时在抵抗匹配噪声的表现上优于传统算法,具有更强的鲁棒性.