基于Normal-Gamma共轭先验优化因子分析的图像去噪算法  

IMAGE DENOISING ALGORITHM BASED ON NORMAL-GAMMA CONJUGATE PRIOR OPTIMIZED FACTOR ANALYSIS

作　　者：邱燕玲[1] 胡建明 Qiu Yanling;Hu Jianming(Luoding Polytechnic,Luoding 527200,Guangdong,China;School of Electrical and Information Engineering,Changsha University of Science and Technology,Changsha 410114,Hunan,China)

机构地区：[1]罗定职业技术学院,广东罗定527200 [2]长沙理工大学电气与信息工程学院,湖南长沙410114

出　　处：《计算机应用与软件》2023年第12期237-242,共6页Computer Applications and Software

基　　金：广东省高职高专云计算与大数据专业委员会2019年度课题(GDYJSKT19-05)。

摘　　要：采用传统因子分析(Fact Analysis,FA)模型进行噪声抑制时存在因子个数难以确定、噪声抑制后图像质量下降等问题,基于贝叶斯决策理论提出一种Normal-Gamma共轭先验优化FA模型的图像去噪算法,利用Normal-Gamma分布对FA加载因子和隐变量的概率分布建模,一方面对不适定噪声抑制问题正则化,另一方面提高模型的稀疏性,增加参数估计稳定性,采用变分贝叶斯期望最大(Variational Bayesian Expectation Maximum,VBEM)算法对模型求解,自动确定因子个数的同时提升噪声抑制性能。基于标准图像数据集的实验结果表明,所提算法在实现噪声抑制的同时较好地保留了图像的边缘和纹理等细节信息,并且能够明显提升低信噪比条件下的图像识别性能。The number of factors is difficult to determine when factor analysis(FA)model is used for noise suppression,and the image quality degrades after noise suppression.Aimed at these problems,based on Bayesian decision theory,an image denoising algorithm based on normal-Gamma conjugate prior optimized FA model is proposed.The normal-Gamma distribution was used to model the probability distribution of FA loading factors and hidden variables.The ill posed noise suppression problem was regularized and the stability of parameter estimation was increased to improve the sparsity of the model.The variable Bayesian expectation maximum(VBEM)algorithm was used to solve the model,which could automatically determine the number of factors and improve the noise suppression performance.Experimental results based on standard image data sets show that the proposed algorithm can achieve noise suppression while retaining the edge and texture details of the image,and can significantly improve the image recognition performance under the condition of low signal-to-noise ratio.

关 键 词：图像去噪 分层贝叶斯 因子分析 共轭先验 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置] TP3[自动化与计算机技术—控制科学与工程]