基于高斯过程的点匹配方法

针对点匹配问题,在已知两个点集对应关系的情况下,通过高斯过程回归计算两个点集的空间几何变换模型.首先,给定高斯过程的协方差函数,通过对训练样本的学习计算协方差函数中的未知参数,得到后验的高斯过程.然后,根据后验的高斯过程预测测试样本的目标值,完成函数映射.该方法在空间几何变换模型的非线性映射能力与问题求解复杂度之间折衷,是一个基于训练样本的自动学习过程.仿真结果表明:该方法具有较好的非线性映射能力,并且只需较少的运算时间.

A point matching algorithm based on Gaussian processes

Presented was a new robust point matching algorithm that can estimate the geometric transformations between two corresponding pointsets via Gaussian processes(GP).The commonly used covariance function was selected and its hyperparameters were adapted by learning through training point set to get the posterior GP.Then the posterior GP was used to make predictions for unseen test cases.The tradeoff between penalty and data-fit in the GP model was automatic.Experiments show that the algorithm had efficient nonlinear mapping capability and practical implementation in both synthetic and real cases.