Compressed sensing image reconstruction via adaptive sparse nonlocal regularization
|Author:||Zha, Zhiyuan1; Liu, Xin2,3; Zhang, Xinggan1;|
1School of Electronic Science and Engineering, Nanjing University, Nanjing 210023, China
2The Center for Machine Vision and Signal Analysis, University of Oulu, Finland
3 School of Electronics and Information Engineering, Xi’an Jiaotong University, Xi’an, Shanxi, China
4Department of Electronic Science and Engineering, Nanjing University, and National Mobile Commun. Research Lab, Southeast University, Nanjing 210023, China
5School of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China
|Online Access:||PDF Full Text (PDF, 7.6 MB)|
|Persistent link:|| http://urn.fi/urn:nbn:fi-fe201902215914
|Publish Date:|| 2017-09-10
Compressed sensing (CS) has been successfully utilized by many computer vision applications. However,the task of signal reconstruction is still challenging, especially when we only have the CS measurements of an image (CS image reconstruction). Compared with the task of traditional image restoration (e.g., image denosing, debluring and inpainting, etc.), CS image reconstruction has partly structure or local features. It is difficult to build a dictionary for CS image reconstruction from itself. Few studies have shown promising reconstruction performance since most of the existing methods employed a fixed set of bases (e.g., wavelets, DCT, and gradient spaces) as the dictionary, which lack the adaptivity to fit image local structures. In this paper, we propose an adaptive sparse nonlocal regularization (ASNR) approach for CS image reconstruction. In ASNR, an effective self-adaptive learning dictionary is used to greatly reduce artifacts and the loss of fine details. The dictionary is compact and learned from the reconstructed image itself rather than natural image dataset. Furthermore, the image sparse nonlocal (or nonlocal self-similarity) priors are integrated into the regularization term, thus ASNR can effectively enhance the quality of the CS image reconstruction. To improve the computational efficiency of the ASNR, the split Bregman iteration based technique is also developed, which can exhibit better convergence performance than iterative shrinkage/thresholding method. Extensive experimental results demonstrate that the proposed ASNR method can effectively reconstruct fine structures and suppress visual artifacts, outperforming state-of-the-art performance in terms of both the PSNR and visual measurements.
The visual computer. International journal of computer graphics
|Pages:||117 - 137|
|Type of Publication:||
A1 Journal article – refereed
|Field of Science:||
113 Computer and information sciences
213 Electronic, automation and communications engineering, electronics
This work was supported by the Natural Science Foundation of China (61462052, 61571220) and the open research fund of National Mobile Commune. Research Lab., Southeast University (No.2015D08).
© Springer-Verlag Berlin Heidelberg 2016. This is a post-peer-review, pre-copyedit version of an article published in Visual computer. The final authenticated version is available online at: http://dx.doi.org/10.1007/s00371-016-1318-9.