Robust reconstruction algorithm for compressed sensing in Gaussian noise environment using orthogonal matching pursuit with partially known support and random subsampling
 Parichat Sermwuthisarn^{1},
 Supatana Auethavekiat^{1}Email author,
 Duangrat Gansawat^{2} and
 Vorapoj Patanavijit^{3}
https://doi.org/10.1186/16876180201234
© Sermwuthisarn et al; licensee Springer. 2012
Received: 2 April 2011
Accepted: 15 February 2012
Published: 15 February 2012
Abstract
The compressed signal in compressed sensing (CS) may be corrupted by noise during transmission. The effect of Gaussian noise can be reduced by averaging, hence a robust reconstruction method using compressed signal ensemble from one compressed signal is proposed. The compressed signal is subsampled for L times to create the ensemble of L compressed signals. Orthogonal matching pursuit with partially known support (OMPPKS) is applied to each signal in the ensemble to reconstruct L noisy outputs. The L noisy outputs are then averaged for denoising. The proposed method in this article is designed for CS reconstruction of image signal. The performance of our proposed method was compared with basis pursuit denoising, Lorentzianbased iterative hard thresholding, OMPPKS and distributed compressed sensing using simultaneously orthogonal matching pursuit. The experimental results of 42 standard test images showed that our proposed method yielded higher peak signaltonoise ratio at low measurement rate and better visual quality in all cases.
Keywords
compressed sensing (CS) orthogonal matching pursuit (OMP) distributed compressed sensing modelbased method1. Introduction
Compressed sensing (CS) is a sampling paradigm that provides the signal compression at a rate significantly below the Nyquist rate [1–3]. It is based on that a sparse or compressible signal can be represented by the fewer number of bases than the one required by Nyquist theorem, when it is mapped to the space with bases incoherent to the bases of the sparse space. The incoherent bases are called the measurement vectors. CS has a wide range of applications including radar imaging [4], DNA microarrays [5], image reconstruction and compression [6–14], etc.
 (1)
Define the square matrix, Ω, as the matrix having measurement vectors as its column vectors.
 (2)
Randomly remove the rows in Ω to make the row dimension of Ω equal to the one of Φ.
 (3)
Set Φ to Ω after row removal.
 (4)
Normalize every column in Φ
The optimization of ℓ_{0} norm which is nonconvex quadratically constrained optimization is NPhard and cannot be solved in practice. There are two major approaches for problem solving: (1) basis pursuit (BP) approach and (2) greedy approach. In BP approach, the ℓ_{0} norm is relaxed to the ℓ_{1} norm [15–17]. The y = Φs condition becomes the minimum ℓ_{2} norm of y  Φs. When Φ satisfies the restricted isometry property (RIP) condition [18], the BP approach is an effective reconstruction approach and does not require the exactness of the sparse signal. However, it requires high computation. In the greedy approach [19, 20], the heuristic rule is used in place of ℓ_{1} optimization. One of the popular heuristic rules is that the nonzero components of s correspond to the coefficients of the random measurement vectors having the high correlation to y. The examples of greedy algorithm are OMP [19], regularized OMP (ROMP) [20], etc. The greedy approach has the benefit of fast reconstruction.
The reconstruction of the noisy compressed measurement signals requires the relaxation of the y  Φs constraint. Most algorithms provide the acceptable bound for the error between y and Φs [17–26]. The error bound is created based on the noise characteristic such as bounded noise, Gaussian noise, finite variance noise, etc. The authors in [17] show that it is possible to use BP and OMP to reconstruct the noisy signals, if the conditions of the sufficient sparsity and the structure of the overcompleted system are met. The sufficient conditions of the error bound in basis pursuit denoising (BPDN) for successful reconstruction in the presence of Gaussian noise is discussed in [21]. In [22], the Danzig selector is used as the reconstruction technique. ℓ_{∞} norm is used in place of ℓ_{2} norm. The authors of [23] propose using weighted myriad estimator in the compression step and Lorentzian norm constraint in place of ℓ_{2} norm minimization in the reconstruction step. It is shown that the algorithm in [23] is applicable for reconstruction in the environment corrupted by either Gaussian or impulsive noise.
OMP is robust to the small Gaussian noise in y due to its ℓ_{2} optimization during parameter estimation. ROMP [20, 26] and compressed sensing matching pursuit (CoSaMP) [24, 26] have the stability guarantee as the ℓ_{1}minimization method and provide the speed as greedy algorithm. In [25], the authors used the mutual coherence of the matrix to analyze the performance of BPDN, OMP, and iterative hard thresholding (ITH) when y was corrupted by Gaussian noise. The equivalent of cost function in BPDN was solved through ITH in [27]. ITH gives faster computation than BPDN but requires very sparse signal. In [28], the reconstruction by Lorentzian norm [23] is achieved by ITH and the algorithm is called Lorentzianbased ITH (LITH). LITH is not only robust to Gaussian noise but also impulsive noise. Since LITH is based on ITH, therefore it requires the signal to be very sparse.
Recently, most researches in CS focus on the structure of sparse signals and creation of modelbased reconstruction algorithms [29–35]. These algorithms utilize the structure of the transformed sparse signal (e.g., wavelettree structure) as the prior information. The modelbased methods are attractive because of their three benefits: (1) the reduction of the number of measurements, (2) the increase in robustness, and (3) the faster reconstruction.
Distributed compressed sensing (DCS) [33, 35, 36] is developed for reconstructing the signals from two or more statistically dependent data sources. Multiple sensors measure signals which are sparse in some bases. There is correlation between sensors. DCS exploits both intra and inter signal correlation structures and rests on the joint sparsity (the concept of the sparsity of the intra signal). The creators of DCS claim that a result from separate sensors is the same when the joint sparsity is used in the reconstruction. Simultaneously OMP (SOMP) is applied to reconstruct the distributed compressed signals. DCSSOMP provides fast computation and robustness. However, in case of the noisy y, the noise may lead to incorrect basis selection. In DCSSOMP reconstruction, if the incorrect basis selection occurs, the incorrect basis will appear in every reconstruction, leading to error that cannot be reduced by averaging method.
In this article, the reconstruction method for Gaussian noise corrupted y is proposed. It utilizes the fact that image signal can be reconstructed from parts of y, instead of an entire y. It creates the member in the ensemble of sampled y by randomly subsampling y. The reconstruction is applied to reconstruct each member in the ensemble. We hypothesize that all randomly subsampled y are corrupted with the noise of the same mean and variance; therefore, we can remove the effect of Gaussian noise by averaging the reconstruction results of the signals in the ensemble. The reconstruction is achieved by OMP with partially known support (OMPPKS) [34]. Our proposed method differs from DCS in that it requires only one y as the input. It is simple and requires no complex parameter adjustment.
2. Background
2.1 Compressed sensing
where x is an Ndimensional nonsparse signal; s is a weighted Ndimensional vector (sparse signal with k nonzero elements), and Ψ is an N × N orthogonal basis matrix.
where e is an Mdimensional noise vector.
2.2 Reconstruction method
where δ_{ k } is the krestricted isometry constant of a matrix Φ. RIP is used to ensure that all subsets of k columns taken from Φ are nearly orthogonal. It should be noted that Φ has more column than rows; thus, Φ cannot be exactly orthogonal [2].
The reconstruction algorithms used in the experiment are BPDN, OMPPKS, LITH, and DCSSOMP. They are described in the following sections.
2.2.1 BPDN
where ε is the error bound.
BPDN is often solved by linear programming. It guarantees a good reconstruction if Φ satisfies RIP condition. However, it has the high computational cost as BP.
2.2.2. OMPPKS
OMPPKS [34] is adapted from the classical OMP [19]. It makes use of the sparse signal structure that some signals are more important than the others and should be set as nonzero components. It has the characteristic of OMP that the requirement of RIP is not as severe as BP's [26]. It has a fast runtime but may fail to reconstruct the signal (lacks of stability). It has the benefit over the classical OMP as it can successfully reconstruct y even when y is very small (very low measurement rate (M/N)). It is different from treebased OMP (TOMP) [30] in that the subsequent bases selection of OMPPKS does not consider the previously selected bases, while TOMP sequentially compares and selects the next good wavelet subtree and the group of related atoms in the wavelet tree.
In this article, sparse signal is in wavelet domain, where the signal in LL subband must be included for successful reconstruction. All components in LL subband are selected as nonzero components without testing for the correlation. The algorithm for OMPPKS when the data are represented in wavelet domain is as follows.
Input:

An M × N measurement matrix, Φ = [φ_{1}, φ_{2}, φ_{3}, ..., φ_{ N }]

The Mdimensional compressed measurement signal, y

The set containing the indexes of the bases in LL subbands, Γ = {γ_{1}, γ_{2}, ..., γ_{Γ}}.

The number of nonzero entries in the sparse signal, k.
Output:

The set containing k indexes of the nonzero element in x, Λ_{ k }= {λ_{ i }}; i = 1,2,...,k.
Procedure:
 (a)Select every bases in LL subband.$\begin{array}{c}t=\left\Gamma \right\hfill \\ {\Lambda}_{t}=\Gamma \hfill \\ {\mathbf{\Phi}}_{t}=\left[{\phi}_{{\gamma}_{1}}\phantom{\rule{0.3em}{0ex}}{\phi}_{{\gamma}_{2}}\phantom{\rule{0.3em}{0ex}}...\phantom{\rule{2.77695pt}{0ex}}{\phi}_{{\gamma}_{t}}\right].\hfill \end{array}$
 (b)Solve the least squared problem to obtain the new reconstructed signal, z_{ t }.${\mathbf{z}}_{t}=arg\phantom{\rule{0.3em}{0ex}}\underset{z}{min}{\u2225\mathbf{y}{\mathbf{\Phi}}_{t}\mathbf{z}\u2225}_{2}$
 (c)Calculate the new approximation, a_{ t }, and find the residual (error, r_{ t }). a_{ t } is the projection of y on the space spanned by Φ_{ t }.$\begin{array}{c}{\mathbf{a}}_{t}={\mathbf{\Phi}}_{t}{\mathbf{z}}_{t}\hfill \\ {\mathbf{r}}_{t}=\mathbf{y}{\mathbf{a}}_{t}.\hfill \end{array}$
 (a)
Increment t by one, and terminate if t > k.
 (b)Find the index, λ_{ t }, of the measurement basis, φ_{ j }, that has the highest correlation to the residual in the previous iteration (r_{t1}).${\lambda}_{t}=arg\underset{j=\left[1,N\right],j\notin {\Lambda}_{t1}}{max}\left\u27e8{\mathbf{r}}_{t1},{\phi}_{j}\u27e9\right.$
If the maximum occurs for multiple bases, select one deterministically.
 (c)Augment the index set and the matrix of the selected basis.$\begin{array}{c}{\Lambda}_{t}={\Lambda}_{t1}\cup \left\{{\lambda}_{t}\right\}\phantom{\rule{0.3em}{0ex}}\mathsf{\text{and}}\hfill \\ {\mathbf{\Phi}}_{t}=\left[{\mathbf{\Phi}}_{t1}\phantom{\rule{1em}{0ex}}{\phi}_{{\lambda}_{t}}\right]\phantom{\rule{0.3em}{0ex}}.\hfill \end{array}$
 (d)Solve the least squared problem to obtain the reconstructed signal, z_{ t }.${\mathbf{z}}_{t}=arg\phantom{\rule{0.3em}{0ex}}\underset{z}{min}\u2225\mathbf{y}{\mathbf{\Phi}}_{t}\mathbf{z}\u22252$
 (e)Calculate the new approximation, a_{ t }, that best describes y. Then, calculate the residual, r_{ t }, of the current approximation.$\begin{array}{c}{\mathbf{a}}_{t}={\mathbf{\Phi}}_{t}{\mathbf{z}}_{t}\hfill \\ {\mathbf{r}}_{t}=\mathbf{y}{\mathbf{a}}_{t}\hfill \end{array}$
 (f)
Go to step (a)
The reconstructed sparse signal, $\widehat{\mathbf{x}}$, has indexes of nonzero components listed in Λ_{ k }. The value of the λ_{ j } th component of $\widehat{\mathbf{x}}$ equals to the j th component of z_{ t }. The termination criterion can be changed from t > k to that r_{t1}is less than the predefined threshold.
2.2.3. LITH
where α is a scale parameter. The algorithm for LITH is as follows.
Input:

An M × N measurement matrix, Φ

The Mdimensional compressed measurement signal, y

The number of nonzero entries in the sparse signal, k.
Output:

The reconstructed signal, x.
 (a)
Set x(0) to zero vector and t to 0.
 (b)At each iteration, x(t + 1) was computed by$\mathbf{x}\left(t+1\right)={\mathbf{H}}_{k}\left(\mathbf{x}\mathsf{\text{(}}t\mathsf{\text{)}}\right)+\mu \mathbf{g}\left(t\right)),$where H_{ k }(a) is the nonlinear operator where the k largest components in a are kept but the remaining components are set to zero. μ is the step size. In this article, g is defined as follows.$\mathbf{g}\left(t\right)={\mathbf{\Phi}}^{T}{\mathbf{W}}_{t}\left(\mathbf{y}\mathbf{\Phi}\mathbf{x}\left(t\right)\right).$W_{ t } is an M × N diagonal matrix. The diagonal element in W_{ t } is defined as${\mathbf{W}}_{t}\left(i,i\right)=\frac{{\alpha}^{2}}{{\alpha}^{2}+{\left({y}_{i}{\mathbf{\Phi}}_{i}^{T}x\left(t\right)\right)}^{2}},i=1,....,M.$The step size is set as$\mu \left(t\right)=\frac{{\u2225{\mathbf{g}}_{k\left(t\right)}\left(t\right)\u2225}_{2}^{2}}{{\u2225{\mathbf{W}}_{t}^{1/2}{\mathbf{\Phi}}_{k\left(t\right)}{\mathbf{g}}_{k\left(t\right)}\left(t\right)\u2225}_{2}^{2}}.$
In case that ${\u2225\mathbf{y}\mathbf{\Phi}\mathbf{x}\left(t+1\right)\u2225}_{L{L}_{2},\alpha}>{\u2225\mathbf{y}\mathbf{\Phi}\mathbf{x}\left(t\right)\u2225}_{L{L}_{2},\alpha},$ μ(t) is set to 0.5μ(t).
 (c)
Terminate when the difference between Φx and y is less than or equal to the predefined error.
LITH is the fast and robust algorithm but it faces the same problem as ITH. It requires that either x must be very sparse or y must be very large (high measurement rate). It is faster than OMP but with less stability.
2.2.4. DCSSOMP
DCS uses the concept of joint sparsity, which is the sparsity of every signal in the ensemble. It is used under the environment that there are a number of y whose original signals (x) are related. It has three models: sparse common component with innovations, common sparse support, and non sparse common component with sparse innovations [31, 33]. In this article, the common sparse support model is used. SOMP [31, 36] is proposed as the reconstruction algorithm. SOMP is adapted from OMP.
DCSSOMP searches for the solution that contains maximum energy in the signal ensemble. Given that the ensemble of y is {y_{ i }}; i = 1,2,...,L. The basis selection criterion in DCSSOMP is changed from ${\lambda}_{t}=arg\underset{j=\left[1,N\right],j\notin {\Lambda}_{t1}}{max}\u27e8{\mathbf{r}}_{t1},{\phi}_{j}\u27e9$ to ${\lambda}_{t}=arg\underset{j=\left[1,N\right],j\notin {\Lambda}_{t1}}{max}{\sum}_{i=1}^{L}\left\u27e8{\mathbf{r}}_{i,t1},{\phi}_{i,j}\u27e9\right,$ where r_{i,t1}is the residual of y_{ i } to the projection of y_{ i } on to the space spanned by Φ_{t1}. The rest of the procedure remains the same as OMP. The indexes of nonzero components in the reconstructed x_{ i } (i = 1, 2, ..., L) are the same, but the value of nonzero components may differ. It should be noted that when L is equal to one, the DCSSOMP is OMP.
3. Proposed method
This section addresses the problem of image reconstruction from Gaussian noise corrupted y. The block processing is applied to reduce the computational cost. Block processing and the vectorization of the wavelet coefficients is described in Section 3.1. The proposed reconstruction process from the ensemble of y is explained in Section 3.2.
3.1 Block processing and the vectorization of the wavelet coefficients
The waveletdomain image in Figure 1b is divided into blocks along its row as shown in Figure 1c. In Figure 1c, the image has eight rows and is divided into eight blocks. The signal can be made sparser by wavelet shrinkage thresholding [37]. All coefficients in LL_{3} subband are preserved. By using the wavelet shrinkage thresholding, we can set most coefficients in the other subbands to zero with little distinct visual degradation. Each row in Figure 1c is considered as the sparse signal for our study.
3.2. Reconstruction
The reconstruction method is divided into three stages: the construction of the ensemble of y, the reconstruction by OMPPKS, and data merging.
3.2.1. Construction of the ensemble of y
Given that there are L different pMdimension signals in the ensemble of y. p is the ratio of the sampled signal's size to the original size. p and L are predefined. The i th signal in the ensemble is denoted by y_{ i }. The algorithm for constructing y_{ i } is as follows.
Input:

An M × N measurement matrix, Φ

The Mdimensional compressed measurement signal, y

The dimension of y_{ i }, β = pM.
Output:

The i th signal in the ensemble, y_{ i }.

The truncated measurement matrix for y_{ i }, Φ_{ i }
 (a)
Create the set of β random integers, R = {r_{1}, r_{2},...,r_{ β }}, having the following properties.
For all j, l ∈ [1, β], r_{ j } ∈ [1, M] and r_{ j } = r_{ l } only if j = l.
 (b)
Construct y_{ i } by setting the j th component of y_{ i } to the r_{ j }th component of y for all j ∈ [1, β].
 (c)
Construct Φ_{ i }, according to the following function.
For all j ∈ [1, β], set the j th row of Φ_{ i } to the r_{ j }th row of Φ.
3.2.2. Reconstruction by OMPPKS

the reconstruction of the signal at low measurement rate (M/N),

fast reconstruction,

independent reconstruction result for each signal in the ensemble.
The first requirement comes from the fact that the reconstruction is performed on the sampled signal which is smaller than y. The RIP is not always guaranteed. The second requirement is necessary because the reconstruction must be performed L times (L is the number of the signal in the ensemble). The third requirement is the result of taking the information from only one signal. By combining every sampled signal, original noisy y will be acquired. In the proposed algorithm, the denoising by averaging is possible when each y_{ i } has the distinct reconstruction result from one another. Since each y_{ i } carries different set of the y's components, its total noise is different. Consequently, the reconstruction on each y_{ i } gives the result having different noise corrupted to each pixel. The noise in each pixel can be reduced by averaging.
Even though the reconstruction is performed on the ensemble of y as DCS, DCSSOMP is not applicable, since it does not meet the third requirement. Any greedy algorithms applied to each y_{ i } meet the second and the third requirements. The measurement rate can be kept low (the first requirement) by including the model into the reconstruction. OMPPKS [34] is chosen in this algorithm, because its requirement for measurement rate is low. The experiment in [34] shows that the requirement of OMPPKS was lower than CoSaMPPKS.
OMPPKS is applied to every y_{ i } in the ensemble and forms L different sparse signals (wavelet coefficient). At the end of this stage, there are L noisy images.
3.2.3. Data merging
4. Experimental results
4.1. Experiment setup
Since the compression step in CS consists mostly of linear operations, Gaussian noise corrupting the signal in the earlier states is approximated as the Gaussian noise corrupting the compressed measurement vector. The state where the noise corrupted the image was not specified; therefore, we simply corrupted the compressed measurement vector by different level of Gaussian noise indicated by its variance (σ^{2}).
The experiment consists of two parts: (1) the evaluation for the required parameters (L and p) of OMPPKS+RS and DCSSOMP in Section 4.2 and (2) the performance evaluation in Section 4.3.
4.2. Evaluation for L and p
Both OMPPKS+RS and DCSSOMP require the ensemble of y. We randomly subsampled y with the algorithm described in Section 3.1 to create the ensemble. First, we investigated for the size of the ensemble (L) and the size of the signal in the ensemble for the optimum performance of OMPPKS+RS and DCSSOMP. The size of the signal in the ensemble was investigated in term of the ratio to the size of y (p).
The line in the graph of Figure 6 was shown in different color to represent p that was varied. The effect of p was more pronounced in OMPPKS+RS than in DCSSOMP. The maximum PSNR in OMPPKS+RS was achieved when p = 0.6 in all cases, while the maximum PSNR in DCSSOMP was achieved with different value of p. When σ^{2} were 0.05, 0.1, 0.15, and 0.2, the optimum p for DCSSOMP were 0.9, 0.6, 0.7, and 0.6, respectively. No trend could be established for optimum p in DCSSOMP.
The xaxis in Figure 6 represents L. When L was changed, the performance of DCSSOMP was almost unchanged. On the other hand, the performance of OMPPKS+RS was better, when L was larger. When then noise was higher, OMPPKS+RS required larger L to achieve the optimum performance. In order to achieve the best performance, OMPPKS+RS required the larger L than DCSSOMP in all cases. In most cases, DCSSOMP and OMPPKS+RS had already converged to their optimum performance at L = 6 and 31, respectively.
The number of p which provided the highest PSNR
M/N  

0.2  0.3  0.4  0.5  0.6  
σ ^{ 2 }  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS 
0.05  0.7  0.7  0.9  0.6  0.7  0.8  0.9  0.6  0.9  0.6 
0.1  0.8  0.6  0.6  0.6  0.9  0.6  0.7  0.7  0.8  0.7 
0.15  0.7  0.6  0.7  0.6  0.6  0.6  0.7  0.6  0.7  0.6 
0.2  0.7  0.6  0.6  0.6  0.7  0.6  0.7  0.6  0.6  0.6 
The number of L at which the converged PSNR was guaranteed
M/N  

0.2  0.3  0.4  0.5  0.6  
σ ^{ 2 }  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS 
0.05  6  21  6  21  6  31  6  31  6  31 
0.1  6  21  6  21  6  31  6  31  6  31 
0.15  6  21  6  21  6  31  6  36  6  31 
0.2  6  21  6  21  6  31  6  36  6  31 
From Table 2 the optimum L for DCSSOMP was always equal to 6; thus, L for DCSSOMP was set to 6 in Section 4.3. In OMPPKS+RS, the optimum L varied from 21 to 36. Out of 20 cases shown in the table, the optimum L was 31 in 10 cases. The optimum L for OMPPKS+RS was set to 31 in Section 4.3.
4.3. Performance evaluation
The performance of OMPPKS+RS was compared with the ones of BPDN, LITH, OMPPKS, and DCSSOMP in this section. BPDN, LITH, and OMPPKS used the single y to reconstruct the result, while OMPPKS+RS and DCSSOMP used the ensemble of y. The error bound of BPDN was set to σ^{2}. The value of α in LITH was set to the optimum value of 0.25 [28].
4.3.1. Evaluation by PSNR
Figure 7 also indicates that the proposed OMPPKS+RS was the most effective reconstruction at small M/N (< 0.4). When M/N = 0.4 or higher, the PSNR acquired by the reconstruction from OMPPKS+RS and DCSSOMP was approximately the same. At σ^{2} = 0.05 and M/N = 0.6, all techniques achieved approximately the same PSNR. However, when the noise was increased, the reconstruction from the signal ensemble (OMPPKS+RS and DCSSOMP) was better than the performance of the reconstruction from one signal (BPDN, LITH, and OMPPKS) in all cases but at M/N = 0.2.
It should be noted that even though LITH was designed for the reconstruction of noisy signal, its performance was the worst in almost all cases. This was due to its requirement of very sparse data (or very high M/N). Its performance was still not converged at M/N = 0.6; however, M/N could not be increased indefinitely. The major benefit of CS is the capability to reconstruct the signal from small y, so the large M/N will eliminate the CS benefit. For example, at the sparsity rate of 0.1, M/N = 0.5 would lead to y with the size of 50% of the original image size. Such large compressed image could be achieved by conventional image compression techniques. Thus, it was rare that M/N could be increased to 0.5 or larger.
Since OMPPKS+RS and OMPPKS used the same reconstruction method, the PSNR difference between OMPPKS+RS and OMPPKS indicated the PSNR improvement by using the ensemble of y. The average PSNR improvement was more than 1 dB in all σ^{2}. With the exception of σ^{2} = 0.05, the PSNR from OMPPKS+RS at M/N = 0.2 was higher than the one from OMPPKS at M/N = 0.6. It indicated that by using the ensemble of signal, OMPPKS+RS required lower M/N to achieve the same performance level of OMPPKS.
4.3.2. Evaluation by visual inspection
4.3.3. Evaluation between OMPPKS+RS and DCSSOMP at optimum L and p
The average PSNR when p and L were set according to Tables 1 and 2, respectively
M/N  

0.2  0.3  0.4  0.5  0.6  
σ ^{ 2 }  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS  DCSSOMP  OMPPKS+RS 
0.05  10.00  19.96  19.51  21.23  21.72  21.72  21.64  22.65  23.39  24.21 
0.1  14.03  18.41  17.31  18.79  19.01  19.55  18.77  19.92  20.97  20.83 
0.15  13.54  17.27  16.63  17.64  17.65  18.65  18.34  17.82  19.93  19.15 
0.2  13.65  16.21  15.20  16.57  16.70  16.44  17.46  16.86  18.31  17.75 
By comparing Figure 12 with Figures 9, 10, and 11, we found that the PSNR of some reconstructed images in Figure 12 was lower than Figures 9, 10, and 11. At σ^{2} = 0.2, the PSNR of the reconstructed Car based on DCSSOMP dropped from 24.61 dB (Figure 9) to 17.07 dB (Figure 12). The reconstructed image was also degraded visually. On the other hand, the reconstructed Car based on OMPPKS+RS at σ^{2} = 0.1 had 2.31 dB lower PSNR but the visual quality was approximately the same. The PSNR and visual quality drop were also found in other images but with less degree (e.g., the reconstruction of Pallons based on DCSSOMP at σ^{2} = 0.2).
The PSNR drop was caused by the variance of the best p among test images. The visual quality of the reconstruction based on OMPPKS+RS was approximately the same but the one based on DCSSOMP dropped drastically in some cases. Consequently, it was possible to use one p for every image in OMPPKS+RS but p must be determined image by image in DCSSOMP.
From the comparison between OMPPKS+RS and DCSSOMP, it could be concluded that though OMPPKS+RS produced the results with less PSNR than DCSSOMP in some cases, the results had better visual quality. Furthermore, the parameter adjustment in OMPPKS+RS was easier.
The reason behind the noise reduction was because the reconstruction based on OMPPKS+RS produced different result for difference signal in the ensemble; therefore, the noise in each pixel could be reduced by averaging the intensity among signals in the ensemble. On the other hand, DCSSOMP tried to find one result for every signal in the ensemble. Because the ensemble came from only one signal; hence, the noise was the same and the noise went directly to the result.
5. Conclusions
This article proposed the robust CS reconstruction algorithm for image with the presence of Gaussian noise. The proposed algorithm, OMPPKS+RS, firstly applied random subsampling to create the ensemble of L sampled signals. Then OMPPKS was used to reconstruct the signal. The Gaussian denoising was performed by averaging the image reconstruction of every signal in the ensemble. The experiment shows that by using the ensemble of signal, the proposed algorithm improved the PSNR of the original OMPPKS by at least 0.34 dB. Moreover, the proposed algorithm was efficient in removing the noise when the compression rate was high (small measurement rate). For future work, we plan to add the impulsive noise model into OMPPKS+RS to develop the reconstruction algorithm that is robust to both impulsive and Gaussian noises.
Appendix 1: Computational costs of OMP, OMPPKS, OMPPKS+RS, and DCSSOMP
The computational cost of the t th iteration in OMP
Step  The number of multiplication  The number ofℓ_{2} optimization 

(1)${\lambda}_{t}={arg\; max}_{j\notin {\Lambda}_{t}}\phantom{\rule{0.3em}{0ex}}\left\u27e8{\mathbf{r}}_{t1},{\phi}_{j}\u27e9\right.$  M(Nt+1)   
(2) a_{ t } = Φ_{ t }z_{ t }  Mt   
(3) z_{ t } = arg min_{ z }yΦ_{ t }z_{t1}_{2}    ℓ_{2} optimization for t variables 
Total  MN+M  ℓ_{2} optimization for t variables 
The computation cost of the basis preselection in OMPPKS
Step  The number of multiplication  The number ofℓ_{ 2 }optimization 

(1) z_{ t } = arg min_{ z }yΦ_{ t }z_{t1}_{2}    ℓ_{2} optimization for Γ variables 
(2) a_{ t } = Φ_{ t }z_{ t }  Γ   
Total  Γ  ℓ_{2} optimization for Γ variables 
 (1)
The number of multiplication of the first Γth loops is reduced from (MN+M)Γ to Γ.
 (2)
The ℓ_{2} optimization in the first (Γ  1) iterations is removed.
The computation cost of the t th iteration in DCSSOMP
Step  The number of multiplication  The number ofℓ_{ 2 }optimization 

(1) ${\lambda}_{t}={arg\; max}_{j=1,...,N}\phantom{\rule{0.3em}{0ex}}\sum _{l=1}^{L}\left\u27e8{\mathbf{r}}_{l,t1},{\phi}_{j}\u27e9\right.$  LpM(Nt+1)   
(2) a_{ t } = Φ_{ t }z_{ t }  LpMt   
(3) z_{ t } = arg min_{ z }yΦ_{ t }z_{t1}_{2}    L(ℓ_{2} optimization for t variables) 
Total  Lp(MN+M)  L(ℓ_{2} optimization for t variables) 
From (15) to (18), it can be concluded that the computational cost of OMPPKS+RS is approximately pL times the cost of OMPPKS. From (13), (14), (19) and (20), it can be concluded that the computational cost of DCSSOMP is pL times the cost of OMP. Since both OMPPKS+RS and DCSSOMP reconstruct the ensemble of signals, their computational costs are higher than OMP and OMPPKS.
From (17) to (20), it can be concluded that at the same L and p, the cost of OMPPKS+RS is lower than DCSSOMP because of the usage of OMPPKS. However, it was found that the optimum L and p in OMPPKS+RS and DCSSOMP were different. The product of pL was much higher in OMPPKS+RS, so OMPPKS+RS had the highest computational cost. The effect of higher computing cost in OMPPKS+RS can be reduced by parallel processing, because the reconstruction of each signal in OMPPKS+RS can be done separately.
The total computational cost of the reconstruction of a ksparse signal by OMP, OMPPKS, OMPPKS+RS, and DCSSOMP
Method  The number of multiplication  The number ofℓ_{2}optimization 

OMP  (MN + M)k  $\sum _{t=1}^{k}\left({\ell}_{\mathsf{\text{2}}}\mathsf{\text{optimizationfor}}t\phantom{\rule{0.3em}{0ex}}\mathsf{\text{variables}}\right)$ 
OMPPKS  (MN + M)(kΓ) + Γ  $\sum _{t=\Gamma }^{k}\left({\ell}_{\mathsf{\text{2}}}\mathsf{\text{optimizationfor}}t\mathsf{\text{variables}}\right)$ 
OMPPKS+RS  L[p(MN + M)(k  Γ) + Γ]  $L\sum _{t=\Gamma }^{k}\left({\ell}_{\mathsf{\text{2}}}\mathsf{\text{optimizationfor}}t\phantom{\rule{0.3em}{0ex}}\mathsf{\text{variables}}\right)$ 
DCSSOMP  Lp[(MN + M)k]  $L\sum _{t=1}^{k}\left({\ell}_{\mathsf{\text{2}}}\mathsf{\text{optimizationfor}}t\phantom{\rule{0.3em}{0ex}}\mathsf{\text{variables}}\right)$ 
Declarations
Acknowledgements
The authors would like to thank the reviewers for their comments and suggestions. This research has financially been supported by the National Telecommunications Commission Fund (Grant No. PHD/006/2551 to P. Sermwuthisarn and S. Auethavekiat), the Telecommunications Research Industrial and Development Institute (TRIDI).
Authors’ Affiliations
References
 Donoho DL: Compressive sensing. IEEE Trans Inf Theory 2006, 52(4):12891306.MathSciNetView ArticleMATHGoogle Scholar
 Candes EJ, Wakin MB: An introduction to compressive sampling. IEEE Signal Process Mag 2008, 25: 2130.View ArticleGoogle Scholar
 Candes EJ, Romberg J: Sparsity and incoherence in compressive sampling. Inverse Problem 2007, 23(3):969985. 10.1088/02665611/23/3/008MathSciNetView ArticleMATHGoogle Scholar
 Patel VM, Easley GR, Healy DM, Chellappa R: Compressed synthetic aperture radar. IEEE J Sel Topics Signal Process 2010, 4(2):244254.View ArticleGoogle Scholar
 Parvaresh F, Vikalo H, Misra S, Hassibi B: Recovering sparse signal using measurement matrices in compressed DNA microarrays. IEEE J Sel Topics Signal Process 2008, 2(3):275285.View ArticleGoogle Scholar
 Shi G, Gao D, Liu D, Wang L: High resolution image reconstruction: a new imager via movable random exposure. Proc ICIP, Cairo, Egypt 2009, 11771180.Google Scholar
 Yang J, Zhang Y, Yin W: A fast alternative direction method for TVL1L2 signal reconstruction from partial fourier data. IEEE J Sel Topics Signal Proces 2010, 4(2):288297.View ArticleGoogle Scholar
 Marcia RF, Willett RM: Compressive coded aperture superresoltion image reconstruction. In Proc ICASSP. Nevada, U.S.A.; 2008:833836.Google Scholar
 Gan L: Block compressed sensing of natural images. In Proc DSP. Wales, U.K.; 2007:403406.Google Scholar
 Yang Y, Au OC, Fang L, Wen X, Tang W: Perceptual compressive sensing for image signals. In Proc ICME. New York, U.S.A.; 2009:8992.Google Scholar
 Do TT, Tran TD, Gan L: Fast compressive sampling with structurally random matrices. In Proc ICASSP. Nevada, U.S.A.; 2008:33693372.Google Scholar
 Gan L, Do TT, Tran TD: Fast compressive image using scrambled block hadamard ensemble. In Proc EUSIPCO. Lausanne, Switzerland; 2008.Google Scholar
 Mun S, Fowler JE: Block compressed sensing of images using directional transforms. In Proc ICIP. Cairo, Egypt; 2009:30213024.Google Scholar
 Sermwuthisarn P, Auethavekiat S, Patanavijit V: A fast image recovery using compressive sensing technique with block based orthogonal matching pursuit. In Proc ISPACS. Kanazawa, Japan; 2009:212215.Google Scholar
 Chen S, Donoho DL, Saunders M: Atomic decomposition by basis pursuit. SIAM Rev 2001, 43(1):129159. 10.1137/S003614450037906XMathSciNetView ArticleMATHGoogle Scholar
 Donoho DL, Elad M, Temlyakov VN: Stable recovery of sparse overcomplete representations in the presence of noise. IEEE Trans Inf Theory 2006, 52(1):618.MathSciNetView ArticleMATHGoogle Scholar
 Tropp JA: Just relax: convex programming methods for identifying sparse signals in noise. IEEE Trans Inf Theory 2006, 52(3):10301051.MathSciNetView ArticleMATHGoogle Scholar
 Candes EJ, Romberg J, Tao T: Stable signal recovery from incomplete and inaccurate measurements. Commun Pure Appl Math 2006, 59(8):12071223. 10.1002/cpa.20124MathSciNetView ArticleMATHGoogle Scholar
 Tropp JA, Gilbert AC: Signal recovery from random measurements via orthogonal matching pursuit. IEEE Trans Inf Theory 2007, 53(12):46554666.MathSciNetView ArticleMATHGoogle Scholar
 Needell D, Vershynin R: Uniform uncertaintity principle and signal recovery via regularized orthogonal matching pursuit. Found Comput Math 2009, 9(3):317334. 10.1007/s1020800890313MathSciNetView ArticleMATHGoogle Scholar
 Omidiran D, Wainwright MJ: Highdimensional subset recovery in noise: sparsified measurements without loss of statistical efficiency, in Technical report 753. Department of Statistics, UC Berkeley, U.S.A.; 2008.Google Scholar
 Candès EJ, Tao T: The dantzig selector: statistical estimation when p is much larger than n . Ann Stat 2007, 35(6):23132351. 10.1214/009053606000001523View ArticleMathSciNetMATHGoogle Scholar
 Carrillo RE, Barner KE, Aysal TC: Robust sampling and reconstruction methods for sparse signals in the presence of impulsive noise. IEEE J Sel Topics Signal Proc 2010, 4(2):392408.View ArticleGoogle Scholar
 Needell D, Tropp JA: CoSaMP: iterative signal recovery from incomplete and inaccurate samples. Appl Comput Harmonic Anal 2008, 26(3):301321.MathSciNetView ArticleMATHGoogle Scholar
 BenHaim Z, Eldar YC, Elad M: Coherencebased nearoracle performance guarantees for sparse estimation under Gaussian noise. Proc ICASSP, Prague, Czech Republic 2010, 35903593.Google Scholar
 Needell D: Topics in compressed sensing. In Ph.D. Dissertation, Math. Univ. of California, Davis; 2009.Google Scholar
 Blumensath T, Davies ME: Normalized iterative hard thresholding: guaranteed stability and performance. IEEE J Sel Topics Signal Process 2010, 4(2):298309.View ArticleGoogle Scholar
 Carrillo RE, Barner KE: Lorentzian based iterative hard thresholding for compressed sensing, in Proc . IEEE ICASSP, Prague, Czech Republic 2011, 36643667.Google Scholar
 La C, Do MN: Signal reconstruction using sparse tree representation. In Proc of SPIE Conf on Wavelet Applications in Signal and Image Proc. San Diego, U.S.A; 2005:5914.Google Scholar
 La C, Do MN: Treebased orthogonal matching pursuit algorithm for signal reconstruction. In Proc IEEE ICIP. Georgia, U.S.A; 2006:12771280.Google Scholar
 Duarte MF: Compressed sensing for signal ensembles. In Ph.D. Dissertation. Department of Electrical Engineering, Rice University, Houston, TX; 2009.Google Scholar
 He L, Carin L: Exploiting strucrure in waveletbased Baysian compressive sensing. IEEE Trans Signal Process 2009, 57: 34883497.MathSciNetView ArticleGoogle Scholar
 Baron D, Wakin MB, Duarte MF, Sarvotham S, Baraniuk RG: Distributed compressed sensing. In Technical Report, TREE0612. Rice University, Department of Electrical and Computer Engineering, Houston, TX; 2006.Google Scholar
 Carrillo RE, Polania LF, Barner KE: Iterative algorithm for compressed sensing with partially know support. In Proc IEEE ICASSP. Texas, U.S.A; 2010:36543657.Google Scholar
 Xu M, Lu J: Kclustervalues compressive sensing for imaging. EURASIP J Adv Signal Process 2011. doi:10.1186/16876180201175Google Scholar
 Aravind NV, Abhinandan K, Vineeth VA, Suman DS: Comparison of OMP and SOMP in the reconstruction of compressively sensed hyperspectral images. In Proc IEEE ICCSP. Hamirpur, India; 2011:188192.Google Scholar
 Donoho D: Denoising by soft thresholding. IEEE Trans Inf Theory 1995, 38(2):613627.MathSciNetView ArticleMATHGoogle Scholar
 Catté F, Dibos F, Koepfler G: A morphological scheme for mean curvature motion and applications to anisotropic diffusion and motion of level sets. SIAM J Numer Anal 1995, 32(6):18451909.View ArticleMathSciNetMATHGoogle Scholar
 Yaroslavsky LP: Digital Picture ProcessingAn Introduction. Springer Verlag, Berlin, Heidelberg; 1985.View ArticleGoogle Scholar
 Yaroslavsky L, Eden M: Fundamentals of Digital Optics. Birkhauser, Boston; 1996.View ArticleMATHGoogle Scholar
 Coifman RR, Donoho D: TranslationInvariant DeNoising, Wavelets and Statistics. Springer Verlag, New York; 1995:125150.View ArticleMATHGoogle Scholar
 Ordentlich E, Seroussi G, Verdú S, Weinberger M, Weissman T: A discrete universal denoiser and its application to binary image. In Proc IEEE ICIP. Volume 1. Catalonia, Spain; 2003:117120.Google Scholar
Copyright
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.