摘要
稀疏表示图像去噪方法中噪声方差,需事先假定,且K-SVD(k-means singular value decomposition)字典学习方法难以解决参数自动选择问题。为此,由于图像在梯度域的稀疏性优于空间域,提出一种梯度域非参数贝叶斯字典学习图像去噪方法。考虑非参数贝叶斯字典学习图像去噪模型是一个多变量耦合问题,难以求解,利用Bregman和交替迭代方法把该问题分解为多个子问题,利用最小二乘方法和BPFA(beta process factor analysis)字典学习方法分别求解这些子问题,滤除图像噪声,保留原图像的有用信息。实验结果表明,提出方法较GradDLRec算法的峰值信噪比平均可提高1.4 dB左右,重建的图像细节信息更丰富,且该方法具有良好收敛性。
The noise variance in the sparse representation image denoising method needs to be assumed in advance and the K-SVD(k-means singular value decomposition)dictionary learning method is difficult to solve the problem of parameter automatic selection.Since the image sparsity in the gradient domain is better than that of spatial domain,a method of nonparametric Bayesian dictionary learning in gradient domain for image denoising was proposed.Considering that the model of nonparametric Bayesian dictionary learning for image denoising is a multivariable coupling problem which is difficult to solve,the Bregman and alternating iterative methods were used to decompose the model into three sub-problems.After that,these sub problems were solved respectively using the least square method and the BPFA(beta process factor analysis)dictionary learning method,the image noise was filtered out and the useful information of the original image was reserved.The results of experiment show that the peak signal to noise ratio(PSNR)of the proposed method improves approximately 1.4 dB compared with that of the GradDLRec algorithm,and the reconstructed images have more detail information.The proposed algorithm can also converge after certain times of iterations.
作者
朱路
刘松
曹赛男
刘媛媛
ZHU Lu;LIU Song;CAO Sai-nan;LIU Yuan-yuan(College of Information Engineering,East China Jiaotong University,Nanchang 330013,China)
出处
《计算机工程与设计》
北大核心
2020年第3期802-807,共6页
Computer Engineering and Design
基金
江西省杰出青年人才资助计划基金项目(20171BCB23062)
江西省自然科学基金项目(20171BAB204022)
江西省教育厅科学技术研究重点基金项目(GJJ170360)。
关键词
图像去噪
非参数贝叶斯
字典学习
交替迭代
参数选择
image denoising
nonparametric Bayesian
dictionary learning
alternate iteration
parameter selection
