针对图像的概率分布密度函数的不确定,利用有限高斯混合模型逼近图像的概率分布密度函数。理论上证明了有限高斯混合模型可以以任意精度正逼近实数上的非负黎曼可积函数,特别可以逼近任意的概率分布密度函数。实例表明有限高斯混合模型逼近已知分布密度函数或未知分布密度函数时,具有逼近精度高等优点,为函数逼近提供了理论和技术支持。
With the analysis of probability density functions,it can be approximated using Gaussian mixture model. Nonnegative Riemann integrable function in R space, especially probability density function, can be approximated with arbitrary precision by finite sum of Gaussian density function with different parameters. This can be proved to be effective. The computational examples show that this approach can obtain high accuracy such that it can provide new theoretical and technical supports for function approximation.