基于Laplace积核的多变量概率函数估计

Estimation of Multivariable Probability Function Based on Laplace Product Kernel

在统计模式识别方法中,经常涉及到的问题是多变量概率函数的估计。这类问题往往需要大量的样本数据,而且计算量相当大。提出了一种基于Laplace积核的多变量概率函数估计方法,给出了其详细的推导;提出期望最大化迭代算法对所估计函数参数进行求解。最后以道路交通标志的统计识别方法为应用背景,用VC++实现了此估计方法。实验表明本文所提出的估计方法具有收敛速度快,估计误差小的特点。在模式识别和数据拟合中都有应用价值。

In the model of Bayes classification, parameters estimation of muhivariable probability function is the key problem. Solving these problems need large amount of sample data and time-consuming computation. In this paper, we presented a method for estimation of muhivariable probability function based on Laplace product kernel. The mathematic deduction details are given, and Expectation-Maximization （EM） algorithm is adopted in parameters estimation of probability function. At last, we implemented the methodrelated algorithm. Experiments indicate the proposed method have rapid convergence rate and small error. It has significant application in pattern recognition and data fitting.