基于连分式的广义高斯模型UDCT贝叶斯图像去噪  

Bayesian image denoising by generalized Gaussian model based on UDCT and continued fraction

作　　者：杨兴明[1] 牛坡礼 

机构地区：[1]合肥工业大学计算机与信息学院,安徽合肥230009

出　　处：《合肥工业大学学报（自然科学版）》2015年第1期50-54,共5页Journal of Hefei University of Technology：Natural Science

基　　金：安徽省自然科学基金资助项目(090412041)

摘　　要：文章通过研究均匀离散曲波变换(uniform discrete curvelet transform,UDCT)系数统计特性,发现该变换域的系数具有良好的相关性,且能有效解决广义高斯模型的参数拟合问题。在利用广义高斯模型的参数估计进行图像去噪过程中,从矩估计和最大似然估计出发,采用比牛顿迭代法更稳定的连分式迭代法来求解最大似然估计的超越方程;采用蒙特卡洛方法代替鲁棒中值法来精确地估计每个子带的噪声方差;在Bayesian最大后验概率估计的框架下完成图像去噪。实验结果表明,文中提到的算法与传统的VisuShrink、BayesShrink和SureShrink相比,具有较好的去噪效果和峰值信噪比。Based on the researches on statistical properties of the coefficients of uniform discrete curvelet transform（UDCT）,it is discovered that the coefficients of transform domain have good correlation,and can be used to solve the problem of parameter fitting in generalized Gaussian model effectively.Firstly,in view of the moment estimation and maximum likelihood estimation,the continued fraction iteration method is used,which is more stable than Newton iteration method,to make the maximum likelihood estimation of transcendental equation in the process of image denoising by using the generalized Gaussian model parameter estimation.Then the Monte Carlo method is used to estimate each subband noise variance accurately instead of robust median method.Finally,the image denoising is completed in the framework of Bayesian maximum posterior probability estimation.The experimental results show that the proposed algorithm has good denoising effect and peak signal-to-noise ratio（SNR）in comparison with traditional methods such as VisuShrink,BayesShrink and SureShrink.

关 键 词：广义高斯模型 连分式迭代法 均匀离散曲波变换 蒙特卡洛方法 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]