摘要
组稀疏学习在图像去噪中显示出巨大的潜力,但现有方法仅从图像块级别考虑含噪图像的非局部自相似性,影响了强噪声图像的重建质量.文中在组稀疏复原模型中引入组稀疏残差和全变分正则化约束,将含噪图像复原问题转化为多尺度图像块匹配和减小组稀疏残差;基于干净图像的组稀疏系数预估和多尺度图像块匹配,提出了自适应图像复原迭代算法,以提升组稀疏学习算法的图像去噪和精细结构复原能力.实验结果表明,文中算法能更好地保留图像的细节纹理,减少过平滑和伪影现象,在强噪声图像复原的主、客观综合评价上优于BM3D、WNNM等标杆去噪算法.
Compressed sensing based on group sparsity has shown gloat potential in image denoising. However, most existing methods considm-d nonlocal self-similarity prior of noisy images only in a block-wise manner, which reduced reconstruction quality. This paper introduces group sparsity residual and total variance as the rsgularization constraints into image rsstoration model based on group sparsity and transfomls the reconstruction problem into muhiscale patch matching and decreasing group sparsity residual. Then, a self-adaptive iterative algorithm for image restoration is proposed based on estimating original images＇ group sparee coefficients and matching patches at multiple scales, which improve group sparsity learning＇s performance in denoising and restoring fine strutting. Experimental results denmnstrate that the proposed algorithm can retain more details, reduce over-smoothness and artifacts. The proposed algorithm outperforms the contrast benchmarking algorithms for images cmwupted with strong noise, such as BM3D, WNNM when considering the visual results and the objective evaluation together.
作者
高红霞
陈展鸿
曾润浩
罗澜
陈安
马鸽
GAO Hongxia1, CHEN Zhanhong1 ,ZENG Runhao1 ,LUO Lan1 ,CHEN An1 ,MA Ge2(1.School of Automation Science and Engineering, South China University of Technology, Guangzhou 510640, Guangdong, China; 2. School of Mechanical and Electric Engineering, Guangzhou University, Guangzhou 510006, Guangdong, China)
出处
《华南理工大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2018年第8期11-18,共8页
Journal of South China University of Technology(Natural Science Edition)
基金
国家自然科学基金资助项目(61403146
61603105)
广州市科技计划项目(201707010054
201704030072)
华南理工大学中央高校基本科研业务费专项资金资助项目(2015ZM128)~~
关键词
图像去噪
强噪声图像
组稀疏残差
自适应正则化算法
非局部自相似性
多尺度图像块匹配
image denoising
images corrupted with strong noise
group sparsity residual
self-adaptive regulari-zation algorithm
nonlocal self-similarity
muhiscale patch matching
