CAJ | 학술논문

在概率图模型框架下提出了一种将回归分析和聚类分析相结合的贝叶斯点集匹配方法，其中，回归分析用来估计两个点集之间的映射函数，而聚类分析用来建立两个点集中点与点之间的对应关系。本文将点集匹配问题表示为一种多层的概率有向图，并提出了一种由粗到精的变分逼近算法来估计点集匹配的不确定性；此外，还利用高斯混合模型估计映射函数回归中的异方差噪声和场景点密度估计中离群点的分布；同时，引入转移变量建立起模型点集与场景点集之间的关系，并与离群点混合模型共同对场景点的分布进行估计。实验结果表明，该方法与其他点集匹配算法相比，在鲁棒性和匹配精度方面均达到了较好的效果。
재개솔도모형광가하제출료일충장회귀분석화취류분석상결합적패협사점집필배방법，기중，회귀분석용래고계량개점집지간적영사함수，이취류분석용래건립량개점집중점여점지간적대응관계。본문장점집필배문제표시위일충다층적개솔유향도，병제출료일충유조도정적변분핍근산법래고계점집필배적불학정성；차외，환이용고사혼합모형고계영사함수회귀중적이방차조성화장경점밀도고계중리군점적분포；동시，인입전이변량건립기모형점집여장경점집지간적관계，병여리군점혼합모형공동대장경점적분포진행고계。실험결과표명，해방법여기타점집필배산법상비，재로봉성화필배정도방면균체도료교호적효과。
In this work, we propose a combinative strategy of regression and clustering for point set matching problems under Bayesian framework, with regression for estimation of transformation and clustering for establishment of correspon-dence. The structure of matching problem is represented by a hierarchical graph model and the matching uncertainty is approximated by a coarse-to-fine variational inference algorithm. Furthermore, Gaussian mixture models are proposed for the density estimation of heteroscedastic regression noise and spurious outliers in the scene, and the isotropic or anisotropic covariance is imposed on each individual mixture component in terms of the transformed model points. Experimental results show that the proposed approach achieves comparable performance to the state-of-the-art matching algorithms in both robustness and accuracy.