Digital Signal Processing Reference
In-Depth Information
The Application of Wavelet-Based Contourlet Transform
on Compressed Sensing
Mei Du
1,2,3,4,*
, Huaici Zhao
1,3,4
, Chunyang Zhao
1,2,3,4
, and Bo Li
1,2,3,4
1
Department of Optical-Electronics and Information Processing,
Shenyang Institute of Automation, Chinese Academy of Science,
Shenyang 110016, China
2
Graduate School of Chinese Academy of Science,
Beijing 100049, China
3
Key Laboratory of Optical-Electronics Information Processing,
Chinese Academy of Science,
Shenyang 110016, China
4
Key Laboratory of Image Understanding and Computer Vision,
Liaoning Province, Shenyang 110016, China
dumei@sia.cn
Abstract.
Reasonable sparse representation of signals are one of the key factors
to ensure the quality of compressed sampling, so a proper sparse representing
methods should be selected to make the signals sparse to the greatest extent in
the applications of compressed sensing. In this paper we adopted the frame-
work of block compressed sensing to sample the images, used the iterative hard
thresholding(IHT) algorithm to reconstruct the original images, and employed
the wavelet-based contourlet transform, an improved contourlet transform, to
decompose 2D images in IHT reconstruction process. Numerical experiments
indicated that the runtime of the reconstruction algorithm adopting wavelet-
based contourlet transform is the shortest compared to that adopting contourlet
transform and that adopting wavelet transform; under low compression ratios,
the quality of the reconstructed images using wavelet-based contourlet trans-
form is superior to that using contourlet transform and that using traditional
wavelet transform.
Keywords:
Sparse Representation, Wavelet-Based Contourlet Transform,
Block Compressed Sensing, Iterative Hard Thresholding Algorithm.
1
Introduction
Recently an innovative idea named Compressed Sensing appeared in signal
processing area[1,2]. The central idea of compressed sensing(CS) is that, when deal-
ing with signals which are highly compressible in some transform domain, one can
take much less samples than traditional ones which are contributing of the whole data
stream. There are some key factors in CS and selecting a proper sparse basis is one
*
Corresponding author.
