WCSE

Compressed Sensing Network based on Wavelet Transform

Abstract— The traditional compressed sensing algorithm realizes the optimization of image reconstruction through multiple iterative calculation from limited measurements, which cost high computational complexity and long reconstruction time. As the development of deep learning, it is proposed to combine the technology with compressed sensing(CS) which shows great advantages in accurate and fast CS reconstruction. In this paper, we propose a novel algorithm synthesize the advantages of the two technology as well as add another sparse prior technique based sym8 wavelet, which dubbed WCS-Net, is focus on two parts: sampling network based on sparse representation and deep reconstruction elastic network. Experimental results show the WCS-Net has the advanced performance at measurement rates 0.01, 0.04, 0.1, and 0.25, respectively, while maintaining the same running speed as existing image compression methods based on deep learning

Index Terms— compressed sensing; sampling network; sparse representation; deep learning

Zhu Yin
Hefei Institutes of Physical Science, Chinese Academy of Sciences, CHINA
University of Science and Technology of China, CHINA
ZhongCheng Wu
Hefei Institutes of Physical Science, Chinese Academy of Sciences, CHINA
University of Science and Technology of China, CHINA
Jun Zhang
Hefei Institutes of Physical Science, Chinese Academy of Sciences, CHINA
Fang Li
Hefei Institutes of Physical Science, Chinese Academy of Sciences, CHINA

[Download]

Cite: Zhu Yin, ZhongCheng Wu, Jun Zhang, Fang Li, "Compressed Sensing Network based on Wavelet Transform," 2021 The 11th International Workshop on Computer Science and Engineering (WCSE 2021), pp. 339-344, Shanghai, China, June 19-21, 2021.