模式识别中幂变换正态性的研究

Research on the Normality of Power Transformation in Pattern Recognition

摘要:

在基于模板匹配的模式识别中,当数据呈正态分布时得到的识别效果最优。幂变换是一种将非正态分布转换为正态分布的一种有效方法,文中对幂变换技术的正态性问题进行了相关的研究,并探讨了典型分布条件下的幂变换最佳取值问题,为幂变换的应用提供了一定的理论基础。

Abstract:

In pattern recognition based on template matching, the recognition performance can possess optimality properties when the input raw data is normal or Gaussian distribution. Power transformation can be used to convert the data to a normal-like or normal distribution. In this paper, some researches on the normality of power transformation were studied. And the optimum values of power transformation coefficient for different distribution data were also proposed, which could be advantageous to practical application.