摘要
基于小波分析的贝叶斯(Bayes)图像处理方法常常需要获得图像小波波系数的先验概率分布密度,该文提出,利用混合高斯模型对正交小波域中自然图像的父子小波系数的联合分布密度进行建模,运用非完备数据的极大似然估计算法——期望极大(EM)算法,对该模型的参数进行估计并且给出了联合分布密度函数的模型分量数与迭代次数的确定过程。最后,在后验均值(PM)方法下,把该联合分布密度模型运用于图像去噪研究;仿真结果表明该方法能够获得较好的效果。
In this paper, we present a method to analyse the dependent characters of the wavelet coefficients of natural images. Generally, the relationship between the wavelets coefficients have been ignored, here we use joint pdf to study the relationship between the wavelet coefficients. The joint pdf in wavelets is modeled by mixture Gaussian model and its parameters are estimated by the incomplete data Maximum Likelihood method - EM algorithm. The methods for deciding the number of mixture Gaussian model components and the iterative numbers are given . Finally, An example is given in which we use the joint pdf into PM methods to denoise the i. i. d Gaussian noisy images and we get a good simulation.
出处
《计算机仿真》
CSCD
2005年第6期71-74,共4页
Computer Simulation
