光学 精密工程, 2013, 21 (9): 2405, 网络出版: 2013-09-25   

基于t分布混合模型的点集非刚性配准算法

Point set non-rigid registration using t-distribution mixture model
作者单位
1 中国科学院 长春光学精密机械与物理研究所, 吉林 长春 130033
2 中国科学院大学, 北京 100049
3 中国科学院 苏州生物医学工程技术研究所, 江苏 苏州 215163
4 苏州大学附属第一医院, 江苏 苏州 215006
摘要
考虑高斯混合模型(TMM)的点集非刚性配准算法易受异常点和重尾点的影响, 提出了基于t分布混合模型的运动一致性非刚性配准算法。通过期望最大化(EM)框架的完整数据定义将高斯混合模型推广为t分布混合模型, 使用EM算法最小化参数的条件期望获得非刚性配准参数的闭合解。在EM算法中计算浮动点集各个点的先验权重, 减小异常点和重尾点对配准结果的影响; 计算浮动点集各个点的自由度, 自适应地改变每个点的概率密度分布模型, 提高算法的鲁棒性, 并避免了异常点水平估计误差对配准结果的影响。在t分布混合模型的条件期望函数中加入点集位移的正则项, 使邻近点具有运动一致性(CPD)。仿真数据表明, 当噪声水平很高时, TMM-CPD仍可以精确配准点集, 且误差仅为对比算法的1/10。真实图像的近似椭圆状分布、管状分布和三维点云状分布的点集配准结果表明, TMM-CPD的配准误差仅为对比算法的42.0%、80.1%和77.5%。实验表明, TMM-CPD配准含有重尾点和异常点的点集, 具有精度高、鲁棒性好和受重尾点与异常点干扰小等优点。
Abstract
A robust non-rigid registration approach with a t-distribution Mixture Model(TMM) was proposed because point sets as Gaussian mixture model is vulnerable to the outliers and the data with longer than normal tails. The Gaussian mixture model was extended to the student′s-t Mixture Model by full data definition in the Expectation Maximazation(EM) frame, then the closed solutions of the parameter set of the t-distribution mixture model were solved by re-parameterization of the t-distribution mixture model in the EM algorithm. The priori-weight of each float point was calculated in EM framework to reduce the effects of outliers and the data with longer than normal tails on the matching results. The degree of freedom of each float point in the t-distribution mixture model was calculated to change the probability density distribution, improve the robustness of the algorithm and avoid the effect of estimating the outlier level of point sets that may bring additional errors. The conditional expectation function in t-distribution mixture model was added a regular item of point set, so that the points have a feature of Coherent Point Drift(CPD). The simulation data show that the error from the TMM-CPD is only one tenth of that from comparison algorithms. When the point sets are approximate ellipse shape, tubular and three dimensions, the registration errors of TMM-CPD are only 42.0%, 80.1% and77.5% of those using comparison algorithms, respectively. The experiments show that this non-rigid registration approach using the t-distribution mixture model has features of high-accuracy, good robustness compared to other point set registration algorithms for point sets containing outliers and data with longer than normal tails.

周志勇, 薛维琴, 郑健, 蒯多杰, 张涛, 胡粟. 基于t分布混合模型的点集非刚性配准算法[J]. 光学 精密工程, 2013, 21(9): 2405. ZHOU Zhi-yong, XUE Wei-qin, ZHENG Jian, KUAI Duo-jie, ZHANG Tao, HU Su. Point set non-rigid registration using t-distribution mixture model[J]. Optics and Precision Engineering, 2013, 21(9): 2405.

本文已被 2 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!