通信学报 ›› 2017, Vol. 38 ›› Issue (2): 196-202.doi: 10.11959/j.issn.1000-436x.2017041

• 学术通信 • 上一篇    

加权结构组稀疏表示的图像压缩感知重构

李佳1,高志荣2(),熊承义1,周城1   

  1. 1 中南民族大学电子信息工程学院智能无线通信湖北省重点实验室，湖北 武汉 430074
    2 中南民族大学计算机科学学院，湖北 武汉 430074
  • 出版日期:2017-02-01 发布日期:2017-07-20
  • 作者简介:李佳（1991-），女，湖北孝感人，中南民族大学硕士生，主要研究方向为图像复原、压缩感知。|高志荣（1972-），女，湖北仙桃人，中南民族大学副教授，主要研究方向为图像处理与识别、压缩感知。|熊承义（1969-），男，湖南临澧人，中南民族大学教授、硕士生导师，主要研究方向为图像与视频编码、图像分类与识别。|周城（1979-），男，湖北武汉人，中南民族大学讲师、硕士生导师，主要研究方向为视频编码与通信。
  • 基金资助:
    国家自然科学基金资助项目(61471400)

Image compressive sensing recovery based on weighted structure group sparse representation

Jia LI1,Zhi-rong GAO2(),Cheng-yi XIONG1,Cheng ZHOU1   

  1. 1 Hubei Key Lab of Intelligent Wireless Communication,College of Electronic and Information Engineering,South-Central University for Nationalities,Wuhan 430074,China
    2 College of Computer Science,South-Central University for Nationalities,Wuhan 430074,China
  • Online:2017-02-01 Published:2017-07-20
  • Supported by:
    The National Natural Science Foundation of China(61471400)

在线阅读

12

PDF下载

715

被引次数

4

摘要：

利用图像的非局部相似性先验以提升图像恢复质量已得到广泛关注。为了更有效地提升压缩感知（CS）图像的重构质量，提出了一种基于加权结构组稀疏表示（WSGSR）的图像压缩感知重构方法。采用非局部相似图像块结构组加权稀疏表示的 l 1 范数作为规则化项约束优化重构，实现在更好地恢复图像高频细节信息的同时有效减少对图像低频成分的损失，图像重构质量得到明显改善。推导出一种加权软阈值收缩方法，实现对模型的优化求解，对幅值较大的重要系数采用较小的阈值收缩处理，对幅值较小的非重要系数采用相对较大的阈值收缩处理。实验结果比较验证了所提方法的有效性。

关键词: 压缩感知, 图像重构, 加权结构组稀疏表示, 加权软阈值收缩

Abstract:

Non-local similarity prior has been widely paid attention to efficiently improve image recovery quality.To fur-ther improve the recovered image quality for compressive sensing (CS),an image compressive sensing recovery method based on reweighted structure group sparse representation (WSGSR) was proposed. l 1 -norm of WSGSR of image non-local similar patch group was used as a regularization term to optimize reconstruction,which achieved well reserving image high-frequency detail with less loss of image low-frequency component,and thus considerably improve the recon-structed image quality.A reweighted soft thresholding shrinkage method was deduced to achieve optimization solution,in which the significant coefficient with large magnitude value was shrunk by a small threshold,while the non-significant coefficient with small magnitude value was shrunk by a relative large threshold.Experimental results comparison demon-strate the effectiveness of the proposed method.

Key words: compressive sensing, image reconstruction, weighted structure group sparse representation, weighted soft thresholding shrinkage

中图分类号: 

No Suggested Reading articles found!