Research article

Nonmonotone variable metric Barzilai-Borwein method for composite minimization problem

  • Received: 05 March 2024 Revised: 23 April 2024 Accepted: 29 April 2024 Published: 09 May 2024
  • MSC : 65K10, 90-08, 90C30

  • In this study, we develop a nonmonotone variable metric Barzilai-Borwein method for minimizing the sum of a smooth function and a convex, possibly nondifferentiable, function. At each step, the descent direction is obtained by taking the difference between the minimizer of the scaling proximal function and the current iteration point. An adaptive nonmonotone line search is proposed for determining the step length along this direction. We also show that the limit point of the iterates sequence is a stationary point. Numerical results with parallel magnetic resonance imaging, Poisson, and Cauchy noise deblurring demonstrate the effectiveness of the new algorithm.

    Citation: Xiao Guo, Chuanpei Xu, Zhibin Zhu, Benxin Zhang. Nonmonotone variable metric Barzilai-Borwein method for composite minimization problem[J]. AIMS Mathematics, 2024, 9(6): 16335-16353. doi: 10.3934/math.2024791

    Related Papers:

    [1] Xiaoping Xu, Jinxuan Liu, Wenbo Li, Yuhan Xu, Fuxiao Li . Modified nonmonotonic projection Barzilai-Borwein gradient method for nonnegative matrix factorization. AIMS Mathematics, 2024, 9(8): 22067-22090. doi: 10.3934/math.20241073
    [2] Jamilu Sabi'u, Ibrahim Mohammed Sulaiman, P. Kaelo, Maulana Malik, Saadi Ahmad Kamaruddin . An optimal choice Dai-Liao conjugate gradient algorithm for unconstrained optimization and portfolio selection. AIMS Mathematics, 2024, 9(1): 642-664. doi: 10.3934/math.2024034
    [3] Sani Aji, Aliyu Muhammed Awwal, Ahmadu Bappah Muhammadu, Chainarong Khunpanuk, Nuttapol Pakkaranang, Bancha Panyanak . A new spectral method with inertial technique for solving system of nonlinear monotone equations and applications. AIMS Mathematics, 2023, 8(2): 4442-4466. doi: 10.3934/math.2023221
    [4] Jamilu Sabi'u, Ali Althobaiti, Saad Althobaiti, Soubhagya Kumar Sahoo, Thongchai Botmart . A scaled Polak-Ribie`re-Polyak conjugate gradient algorithm for constrained nonlinear systems and motion control. AIMS Mathematics, 2023, 8(2): 4843-4861. doi: 10.3934/math.2023241
    [5] Salwa Syazwani Mahzir, Md Yushalify Misro . Enhancing curve smoothness with whale optimization algorithm in positivity and monotonicity-preserving interpolation. AIMS Mathematics, 2025, 10(3): 6910-6933. doi: 10.3934/math.2025316
    [6] Abdelilah Hakim, Anouar Ben-Loghfyry . A total variable-order variation model for image denoising. AIMS Mathematics, 2019, 4(5): 1320-1335. doi: 10.3934/math.2019.5.1320
    [7] Carlos F. Álvarez, Javier Henríquez-Amador, John Millán G., Eiver Rodríguez . On the composition operator with variable integrability. AIMS Mathematics, 2025, 10(2): 2021-2041. doi: 10.3934/math.2025095
    [8] Yan Ning, Daowei Lu, Anmin Mao . Existence and subharmonicity of solutions for nonsmooth p-Laplacian systems. AIMS Mathematics, 2021, 6(10): 10947-10963. doi: 10.3934/math.2021636
    [9] Faizan A. Khan, Rohit K. Bhardwaj, Tirth Ram, Mohammed A. S. Tom . On approximate vector variational inequalities and vector optimization problem using convexificator. AIMS Mathematics, 2022, 7(10): 18870-18882. doi: 10.3934/math.20221039
    [10] Farah Liyana Azizan, Saratha Sathasivam, Nurshazneem Roslan, Ahmad Deedat Ibrahim . Logic mining with hybridized 3-satisfiability fuzzy logic and harmony search algorithm in Hopfield neural network for Covid-19 death cases. AIMS Mathematics, 2024, 9(2): 3150-3173. doi: 10.3934/math.2024153
  • In this study, we develop a nonmonotone variable metric Barzilai-Borwein method for minimizing the sum of a smooth function and a convex, possibly nondifferentiable, function. At each step, the descent direction is obtained by taking the difference between the minimizer of the scaling proximal function and the current iteration point. An adaptive nonmonotone line search is proposed for determining the step length along this direction. We also show that the limit point of the iterates sequence is a stationary point. Numerical results with parallel magnetic resonance imaging, Poisson, and Cauchy noise deblurring demonstrate the effectiveness of the new algorithm.



    In this paper, we consider the following composite optimization problem:

    minxRnϕ(x)=f(x)+h(x), (1.1)

    where f(x) is a differentiable function with a Lipschitz constant L>0 for the gradient and h(x) is a convex function that is possibly nonsmooth. Many models in sparse optimization can be formulated as problems (1.1), such as the l1-regularized minimization in compressive sensing [1,2,3] and total variation regularization in image processing [4,5,6].

    A special case of model (1.1) is the constrained problem over a convex set, when h(x) is the indicator function of a convex set Ω, i.e.,

    h(x)=lΩ(x)

    and

    lΩ(x)={0,ifxΩ,+,ifxΩ.

    For this case, a simple and famous algorithm is the gradient projection method [7]. To accelerate this projection algorithm, a scheme with line-search along the feasible direction is proposed in [8]. Using the Barzilai-Borwein (BB) stepsize [9] and nonmonotone line search strategy, Birgin et al. [10] developed the nonmonotone spectral projected gradient method, which is particularly appealing for large-scale problems. Dai and Fletcher [11] induced a new projected gradient method that alternately used the two BB steplengths. In [12], the authors proposed a modified spectral conjugate gradient projection algorithm and applied it to the total variation image restoration.

    The other special case of model (1.1), which has attracted much interest in signal and image processing and machine learning, is the well-known l2-l1 problem, i.e.,

    h(x)=1.

    One of the most popular methods is the iterative shrinkage thresholding algorithm [13]. Using Nesterov's technique, Beck and Teboulle [14] proposed the famous fast iterative shrinkage thresholding algorithm (FISTA) by forward-backward splitting. Hale et al. [2] developed the fixed-point continuation method for l1-minimization. They also used BB stepsize and nonmonotone line search to enhance this algorithm in [15]. Another closely related method is the sparse reconstruction by separable approximation algorithm (SpaRSA) [16], which involves minimizing a non-smooth convex problem with separable structures. Hager et al. [17] showed that SpaRSA has a sublinear convergence rate for general convex functions or R-linear when the objective function is strongly convex. An improved version of SpaRSA based on a cyclic BB iteration and adaptive line search was also presented in [17]. Huang and Liu [18] proposed a BB-type method for minimizing composite function. A nonmonotone strategy was employed for determining the step length along the search direction, which was defined by the difference between the minimizer of the approximation objective function and the current iteration point. Recently, Cheng et al. [19] induced an algorithm framework for the more general type of (1.1). At each step, the search direction was obtained by solving an optimization problem. Xiao et al. [20] proposed a nonmonotone BB gradient algorithm for l1-regularized nonsmooth minimization in compressive sensing. At each iteration, the search direction can be easily derived by minimizing a local approximal quadratic model. Tseng and Yun [21] gave a block coordinate gradient descent method. But numerical results showed the efficiency of this method only for small and medium-scale problems.

    There are many variable metric forward-backward (FB) approaches for solving (1.1) in recent literatures. In this type of algorithm, the suitable symmetric positive definite scaling matrices changed at each iteration. A clever choice of them can lead to an improvement in convergence speed [22,23]. Bonettini et al. [24] developed a variable metric inexact-line-search-based method for nonsmooth optimization. An Armijo-like rule was used to determine the stepsize along the descent direction. For the general nonconvex case, they proved that all limit points of the iterate sequence are stationary. In the last two decades, more variants of FB approaches have been proposed. These FB variants introduce a kind of extrapolation, or inertial, step exploiting the two previous iterations and can be traced back to two main typologies: FISTA-like methods [25] and heavy-ball ones [26]. Other recent results can be found in [27,28], where the analysis applies also to the nonconvex case, assuming that the Kurdyka-Lojasiewicz property holds.

    In this paper, following the same lines as in [24] that exploiting some variable metric information and nonmonotone line search in [18,19] can lead to fast image restorations, we propose a BB-type algorithm with adaptive nonmonotone line search and scaling technique. The nonmonotone strategy includes a convex combination of the maximum function value of some previous iterations and the current function value. At each step, the search direction of the algorithm is obtained by solving an optimization subproblem involving a quadratic term with a variable metric matrix and BB steplength plus h(x). Then, we perform a nonmonotone line search along this direction. We prove that the method with the nomonotone line search techniques is globally convergent. Similar to [17,18,24], the new method does not need to know the value of the Lipschitz constant L. Then, we apply this method to total variation image restoration, such as parallel magnetic resonance imaging [29,30], Poisson and Cauchy noise removal [24]. Numerical experiments show that our approach is competitive with several other known methods.

    The rest of the paper is organized as follows: In Section 2, we present the new method for solving the model (1.1). The convergence of the new algorithm is proved in Section 3. In Section 4, we show the results of the numerical experiments. Finally, the conclusion is given in Section 5.

    Throughout this paper, we denote x,y=xTy as the inner product of two vectors x,yRn. denotes the Euclidean norm. The scaled Euclidean norm associated with a symmetric positive definite matrix D is

    xD2=xTDx.

    Given

    μmax>μmin>0,

    M[μmin,μmax] denotes the set of all symmetric positive definite matrix with all eigenvalues contained in the interval [μmin,μmax]. For any

    DM[μmin,μmax],

    we have the result

    μminx2xD2μmaxx2. (1.2)

    In this section, we first introduce the search direction and then introduce a new nonmonotone line search to determine the step length along this direction. At last, the new method is presented.

    Our iteration step for solving (1.1) can be described as follows:

    xk+1=xk+λkdk, (2.1)

    where λk(0,1] is the step length and dkRn is the search direction, which is given by

    dk:=argminvRn{f(xk)Tv+αk2vTDkv+h(xk+v)}, (2.2)

    where αk>0, and Dk is a symmetric positive definite matrix. Let

    x~k=xk+dk,

    from (2.2), we can get that x~k is the minimizer of

    minzRnf(xk)+zxk,f(xk)+αk2zxkDk2+h(z). (2.3)

    In fact, the problem (2.3) can be rewritten as the proximity operator style associated with the convex function h(x), which is defined as

    x~k=proxhαkDk(yk)=argminzRnh(z)+αk2zykDk2, (2.4)

    where

    yk=xk(αkDk)1f(xk).

    So, (2.1) is equivalent to

    xk+1=xk+λk(proxhαkDk(xk(αkDk)1f(xk))xk). (2.5)

    The setting of Dk and αk will affect the performance of the method (2.5). A clever choice of them can lead to significant improvements in the practical convergence speed [31,32]. Firstly, about the metric selection, we assume that

    DkM[μmin,μmax].

    Different problems can choose a different scaling matrix Dk. So, we will discuss how to choose Dk in the section on numerical experiments. Next, we will focus on how to choose the parameter αk. Similar to [17,18,19,20], we choose αk as follows:

    αk=sk1,yk1sk1,sk1, (2.6)

    where

    sk1=xkxk1,yk1=f(xk)f(xk1).

    In fact, this step length was first proposed by Barzilai and Borwein for solving unconstrained problems [9]. Due to its efficiency, the BB approach has received considerable attention. As said in [9], a good property of αk is that it satisfies the quasi-Newton condition

    αk:=argminαRαsk1yk12.

    We take

    αk=min{αmax,max{αmin,sk1,yk1sk1,sk1}}, (2.7)

    where αmax>αmin>0 are fixed constants.

    When the search direction is determined, a suitable stepsize along the decent direction should be found to determine the next iterative point. In the following, we will show how to determine the step length λk. We know that the performance of the BB algorithm benefits from the nonmonotone line search (see [10,11]). The earliest nonmonotone line search technique, which differs from the traditional Armijo line search or the Wolfe-Powell line search, was given by Grippo et al. [33] for smooth, unconstrained optimization. In order to overcome some disadvantages of the nonmonotone line search, Amini et al. introduced a more relaxed nonmonotone strategy that included a convex combination of the largest objective function in some recent past iteration and the current function value [34]. However, they are not directly applied to the nonsmooth problems. In [18], the authors proposed a line search for solving the nonsmooth problem (1.1) as follows:

    ϕ(xk+λdk)ϕ(xl(k))γ2λαkdk2, (2.8)

    where γ(0,1) is a constant and

    ϕ(xl(k))=max0jmin{k,M1}ϕ(xkj), (2.9)

    where M is a fixed integer and l(k) is an integer such that

    kmin{k,M1}l(k)k.

    Motivated by the idea from [18,34], we modified the line search (2.8) and used the following acceptance criterion to determine the step length λk:

    ϕ(xk+λdk)Rk(η)γ2λαkdkDk2, (2.10)

    where

    Rk(η)=ηϕ(xl(k))+(1η)ϕ(xk), (2.11)

    and 0η1. Since

    ϕ(xk)ϕ(xl(k)),

    we have

    ϕ(xk)Rk(η)ϕ(xl(k)). (2.12)

    Given all the above derivations, we now describe the nonmonotone variable metric Barzilai-Borwein type algorithm as Algorithm 1 shows.

    Algorithm 1 Variable metric BB method for solving (1.1).
    Step 0: Given an initial point x0Rn, k:=0, η>0,
      M, αmin, αmax, μmin, μmax, γ(0,1).
    Step 1: Stop if dk=0. Otherwise, continue.
    Step 2: Compute dk via (2.2),
      where αk[αmin,αmax], DkM[μmin,μmax].
    Step 3: Compute λk via (2.10).
    Step 4: Let xk+1=xk+λkdk.
    Step 5: Set k:=k+1 and go to Step 1.

    Algorithm 1 is closely related to the method in [18]. However, compared (2.10) with (2.8), it can be easily found differences. Algorithm 1 use the adaptive nonmonotone line search and the variable metric matrix Dk to get the step length λk. If η=1 and Dk=I, Algorithm 1 is reduced to the algorithms in [18]. Similar to the idea in [24], the variable metric technique is adopted to compute the descent direction, while the paper [24] used the modified Armijo line search to determine the step length λk.

    In this section, we analyze the convergence of Algorithm 1.

    First, we give some useful properties of dk. In the nonsmooth case (1.1), dk is a descent direction for ϕ(x) at xk if ϕ(xk;dk)<0, where ϕ(x;d) is the direction derivative of ϕ(x) at x in the direction dRn and is defined as [35]

    ϕ(x;d)=limt0ϕ(x+td)ϕ(x)t.

    A feasible point x is said to be a stationary point of ϕ(x) if ϕ(x;d)0 for all d. Using the definition of ϕ(x;d) and the property of stationary point, we show that the direction defined by (2.2) is descent if dk0 and if dk=0, xk is a stationary point of ϕ(x) in the following two lemmas. The proof is similar to the proofs of Lemmas 2.2 and 2.3 in [19]. So, we omit it here.

    Lemma 3.1. For any xkRn and dkRn determined by (2.2), we have

    ϕ(xk;dk)f(xk)Tdk+h(xk+dk)h(xk)αk2dkDkTdk.

    Lemma 3.2. xkRn is a stationary point of (1.1) if and only if dk=0.

    Next, we show that Algorithm 1 is well defined.

    Lemma 3.3. Let {xk} be a sequence generated by Algorithm 1. If xk is not a stationary point, then the line search (2.10) is satisfied whenever

    0<λmin{1,(1γ)αminμminL}.

    Moreover, if λk satisfies the inequality (2.10), then

    λkmin{1,ρ(1γ)αminμminL}:=c, (3.1)

    where ρ(0,1).

    Proof. By the inequality (2.12), the Lipschitz continuity of f(x) with constant L, the convexity of h(x), and Lemma 3.1, we can get

    ϕ(xk+λdk)Rk(η)ϕ(xk+λdk)ϕ(xk)=f(xk+λdk)f(xk)+h(xk+λdk)h(xk)λ2L2dk2+λf(xk),dk+λ(h(xk+dk)h(xk))λ2L2dk2λαk2dkDk2. (3.2)

    The use of inequality (2.10) yields

    λ2L2dk2λαk2dkDk2γ2λαkdkDk2.

    Using the assumption (1.2) and Lemma 3.2, we can have

    λ(1γ)αminμminL.

    Now, we give the lower bound of λk. Assume that either λk=1 or the line search (2.10) has failed at least once. Therefore, there is a constant a ρ(0,1) such that

    ϕ(xk+λkρdk)>Rk(η)γ2λkραkdkDk2.

    Combining the inequality (3.2) yields

    λkρLdk2>(1γ)αkdkDk2.

    So,

    λk>ρ(1γ)αminμminL.

    This completes the proof.

    The following theorem implies that every accumulation point of {xk} is a stationary point of (1.1). Moreover, if f(x) is a convex function, Algorithm 1 converges to a global solution of (1.1).

    Theorem 3.1. Let the sequences {xk} and {dk} be generated by Algorithm 1, then

    limkdk=0 (3.3)

    and

    limkϕ(xk)=ϕ¯,

    where ϕ¯ is a constant.

    Proof. Using the definition Rk, the inequalities (2.10) and (2.12), we have

    ϕ(xk+1)Rk(η)γ2λαkdkDk2ϕ(xl(k))γ2λαkdkDk2. (3.4)

    It indicates that

    ϕ(xk+1)ϕ(xl(k)).

    On the other hand, from (2.9), we get

    ϕ(xl(k+1))=max0jmin{k+1,M}ϕ(xk+1j)=max{max1jmin{k+1,M}ϕ(xk+1j),ϕ(xk+1)}max{ϕ(xl(k)),ϕ(xk+1)}=ϕ(xl(k)). (3.5)

    Then the sequence {ϕ(xl(k))} is nonincreasing. Using the same assumption as the one in [19] that ϕ(x) is bounded below, there is a constant ϕ¯ such that

    limkϕ(xl(k))=ϕ¯. (3.6)

    By applying (3.4) with k replaced by l(k)1, we have

    ϕ(xl(k))=ϕ(xl(k)1+λl(k)1dl(k)1)Rl(k)1γ2cαmindl(k)1Dl(k)12ϕ(xl(l(k)1))γ2cαminμmindl(k)12.

    Using (3.6), we obtain

    limkdl(k)1=0.

    By the continuity of ϕ(x), we deduce

    limkϕ(xl(k)1)=limkϕ(xl(k)λl(k)1dl(k)1)=ϕ¯.

    Let k=l(k)j, by induction and using the similar proof of Theorem 3.1 in [19], we can show that the following limits are satisfied for j=1,2,,M,

    limkdl(k)j=0.

    For

    1j=l(k)kM,

    we have

    xl(k)=xk+j=1l(i)kλl(k)jdl(k)j.

    This means that

    limkxkxl(k)=0.

    The facts with the continuity of ϕ(x) and (3.6) indicate that

    limkϕ(xk)=limkϕ(xl(k))=ϕ¯. (3.7)

    Taking the limits of (3.4) and (3.7), we obtain

    limkdk=0.

    From Lemma 3.2, we can also know that any limit point of {xk} is a stationary point of ϕ(x). Therefore, the proof is complete.

    In this section, we give some numerical experiments to illustrate the effectiveness of the proposed method. The first test problem of the form (1.1) is the total variation (TV) of magnetic resonance imaging (MRI). As a special type of (1.1), the parallel magnetic resonance imaging model [36,37] can be written as follows:

    minx12Axb2+τxTV,

    where xCn and n is the total number of pixels in the image. ACm×n is a sample matrix, and bCm is the vector of measured Fourier coefficients. TV is the total variation norm, which was first introduced in [5]. τ>0 is the regularization parameter. Obviously,

    f(x)=12Axb2

    is Lipschitz continuous,

    h(x)=τxTV

    is nonsmooth, and ϕ(x) is a convex function.

    Our new method is named NMBBA, NVMBB, and NVMBBA, respectively, in different parameter settings. We compare it with the famous algorithms SpaRSA [16] and NMBB [18]. In all the experiments, we set

    M=5,γ=0.001,αmin=1010,αmax=1010,μmax=1+1/k2=1/μmin.

    In the first experiment, we use a Poisson mask with an acquisition rate 0.243. The image size is 256×256 (named Data1). The regularization parameter is set as τ=0.0005. In NMBBA, η=0.5, Dk=I. In NVMBB,

    η=1,Dk=1/diag(xk).

    In NVMBBA,

    η=0.7,Dk=1/diag(xk).

    In the second experiment, we use radial mask with an acquisition rate 0.424. The image size is 512×512 (named Data2). The regularization parameter is set as τ=0.00037. In NMBBA, η=0.5, Dk=I. In NVMBB,

    η=1,Dk=1/diag(xk).

    In NVMBBA,

    η=0.9,Dk=1/diag(xk).

    For the subproblem (2.4), we use the FISTA [14] method to solve the minimizing total variation proximal function. The inner iteration number is set to 5. The stopping criterion of the five algorithms is:

    xk+1xkxk+10.0001.

    We use signal-to-noise ratio (SNR) as a means of judging performance, which is defined as follows:

    SNR=20log10xkx^xkx0(dB),

    where x0 is the original image, xk is the recovery, and x^ is its mean value. All algorithms are implemented in the Matlab programming environment (Version R2011a). The experiments are performed on a laptop with 2 GB of RAM, a 2 GHz processor, and the Windows 7 operating system.

    The reconstructed images are shown in Figures 1 and 2. It is seen that all algorithms adequately recovered the image from a small set of data samples. We can also see that NMBB appears to have higher accuracy than other methods for Data1 and NMBBA foe Data2.

    Figure 1.  Original image and reconstructed images of Data1 by different algorithms.
    Figure 2.  Original image and reconstructed images of Data2 by different algorithms.

    This can also be seen in Table 1. From the table, we can see that the proposed method is very efficient in the sense that it reaches a similar SNR value in a shorter time (second) and with a lower iteration number (Iter) compared to the other methods.

    Table 1.  Performance comparison of different methods in MRI reconstruction.
    Data1 Data2
    Algorithms Iter Time(s) SNR(dB) Iter Time(s) SNR(dB)
    SpaSRA 61 73.89 19.3562 28 139.93 28.1334
    NMBB 40 38.79 19.6495 27 132.86 27.1879
    NMBBA 30 28.59 19.4433 21 98.20 28.5226
    NVMBB 22 21.74 19.4899 22 110.91 27.9799
    NVMBBA 22 20.31 19.4899 22 107.48 27.9799

     | Show Table
    DownLoad: CSV

    In Figure 3, we plot the objective function value versus CPU time. From the figure, we can see that, among these five algorithms, the decay rates of SpaRSA are much slower than those of the others.

    Figure 3.  Objective function value versus CPU time for Data1 (a) and Data2 (b).

    The reason may be that it has to solve the subproblem for each trial point on each line search, as said in [18]. NMBBA does not use the variable metric technique. This demonstrates that the convex combination of the maximum function value of some previous iteration and the current function value can improve the convergence speed. If we choose the suitable scaling matrix Dk for NVMBB and NVMBBA, they may converge quickly; for example, see their performance Data1. Otherwise, NVMBB and NVMBBA may be a little slower than NMBBA but also faster than SpaRSA and NMBB. So, a suitable combination of the stepsize αk and the variable metric Dk is important for the practical problems discussed in [31].

    The second test problem is TV Poisson noise image deconvolution. When the image is corrupted by Poisson- type noise, the Kullback-Leibler divergence is often used as the data-fitting term. Let b be the observed blurry and noisy image, then the data discrepancy can be given as

    f(x)=b,lnbHx+g1+Hx+g1b,

    where

    HRn×n

    is a blur matrix, 1 is the vector of all ones, and g is a constant background term. To deal with this ill-posed problem, based on the prior information of the image, TV regularization with a nonnegative constraint is added to control the noise and artifacts, i.e.,

    h(x)=ρxTV+lx0(x),

    where ρ>0 is a regularization parameter. Hence, the Poisson noise image restoration problem is reformulated as a minimization problem

    minxRnϕ(x)=b,lnbHx+g1+Hx+g1b+ρxTV+lx0(x).

    So, Algorithm 1 can be used to solve the convex composite optimization problem.

    In this test, we compare it with the method named VMILA in [24]. In Algorithm 1, the variable metric technique is the same as VMILA, with η=1. This is to say, we use the nonmonotone line search instead of the modified Armijo line search in VMILA to determine the step length λk. So, algorithm 1 is named NONVMILA. The two test images are Micro (128×128) and Phantom (256×256). For the micro image, ρ=0.09 and g=0.5. For the Phantom image, ρ=0.004 and g=10. We set M=3 for Micro and M=9 for Phantom, and the other parameters are the same as VMILA. We also used the FISTA [14] method to minimize the total variation of the proximal function. The inner iteration number is set to 1500. We refer the reader to [24] for more details. Both of the compared methods are stopped when the maximum number of iterations is more than 400. The peak-signal noise ratio (PSNR) is used to evaluate the quality of the restored image.

    The numerical performance of image restoration is recorded in Table 2. We present the CPU times in seconds and the PSNR values of the two algorithms for various blurry observations.

    Table 2.  CPU time and PSNR values of two algorithms for Poisson noise image deblurring.
    VMILA NONVMILA
    Images Time(s) PSNR Time(s) PSNR
    Micro 4.4280 52.7215 3.8330 52.8909
    Phantom 10.2243 24.6938 9.7457 25.9434

     | Show Table
    DownLoad: CSV

    From the table, one can see that the two approaches can produce similar behaviors for the two images. Figure 4 displays the visual results of image deblurring. In terms of visual quality, the difference is not significant. This can also be seen from the PSNR values. Figure 5 reports the relative decrease of the objective function with respect to the optimal value ϕ (when x=x0, x0 is the original image) as a function of the computational time. We observe that NONVMILA requires less time.

    Figure 4.  Original images ((a), (e)), blurred images ((b), (f)) with Poisson noise and images recovered by NMILA ((c) PSNR = 52.7215, (g) PSNR = 24.6938) and NONVMILA ((d) PSNR = 52.8909, (h) PSNR = 25.9434).
    Figure 5.  Relative decrease of the objective functions with respect to the computational time for Mirco (a) and Phantom (b).

    The last problem is Cauchy noise image deconvolution. When the image is corrupted by the Cauchy-type noise, the log function is used as the data fitting term. Let b be the observed blurred and noisy image, then the data discrepancy is

    f(x)=12log(γ2+(Hxb)2),

    where

    HRn×n

    is a blur matrix with the 9×9 testing kernel and the variance 1, while has been set equal to 0.02. Based on the prior information of the image, TV regularization with a nonnegative constraint is added to control the noise and artifacts, i.e.,

    h(x)=ρxTV+lx0(x),

    where ρ>0 is a regularization parameter. Hence, the Cauchy noise image restoration problem is reformulated as

    minxRnψ(x)=12log(γ2+(Hxb)2)+ρxTV+lx0(x).

    So, Algorithm 1 can be used to solve this composite optimization problem.

    In this test, we also compare it with the VMILA type in [38,39]. In Algorithm 1, the variable metric technique is the same as in VMILA, with η=0.9, and is also named NONVMILA. We set M=5 and the other parameters are the same as VMILA. The three test images are the cameraman (256×256), the house (256×256) and the bird (256×256). All methods are stopped when the maximum number of iterations is more than 5000 or

    |ψ(xk)ψ(xk10)|106max{1,|ψ(xk)|}.

    The numerical performance of image restoration is reported in Table 3.

    Table 3.  CPU time and PSNR values of two algorithms for Cauchy noise image deblurring.
    VMILA NONVMILA
    Images Time(s) PSNR Time(s) PSNR
    Cameraman 105.2720 26.2798 48.6810 26.2892
    House 87.3230 30.4242 47.5650 30.4463
    Bird 123.6920 27.1690 55.8790 27.2572

     | Show Table
    DownLoad: CSV

    We present the running time in seconds and PSNR values of the two algorithms. From the table, we observe that the two approaches can get similar behaviors for three images in terms of PSNR. We also find that NONVMILA requires less time than VMILA when the stopping criterion is satisfied. Figure 6 displays the visual results of the deblurred image. From Table 3 and Figure 6, it seems that our approach is very efficient for this tested problem.

    Figure 6.  Original images ((a), (e), (i)), blurred images ((b), (f), (j)) with Poisson noise and images recovered by NMILA ((c) PSNR = 26.2798, (g) PSNR = 30.4242, (k) PSNR = 27.1690) and NONVMILA ((d) PSNR = 26.2892, (h) PSNR = 30.4463, (l) PSNR = 27.2572).

    In this paper, we introduce a variable metric algorithm frame for composite optimization problems. With the nonmonotone line search techniques, the convergence of the method is discussed. With the growing use of nonconvex objective functions in applied fields like image processing and machine learning, the need for numerical methods in a fully nonconvex setting has increased significantly. So, future work will focus on a general nonconvex non-Lipschitz function, including theoretical and numerical analysis. Moreover, there is good potential to improve the convergence results using the conditions about the regularity of the cost function and the boundedness of the Hessian matrix [40]. Also, a multi-step inertial forward-backward splitting algorithm will be considered in the future [41,42]. At last, it is interesting to develop a new scaling matrix and BB stepsize for solving the composite optimization problem.

    Xiao Guo: writing-original draft, writing-review and editing, software. Chuanpei Xu: resources, writing-review and editing, supervision. Zhibin Zhu: methodology, writing-review and editing, supervision. Benxin Zhang: resources, writing-review and editing.

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    The authors wish to thank the editor and anonymous referees for their constructive suggestions and comments, which greatly improved the presentation of this paper. This work was supported in part by the National Natural Science Foundation of China under Grants 11901137 and 62171147, in part by the Guangxi Natural Science Foundation under Grant 2024GXNSFAA010512, in part by the Guangxi Key Laboratory of Automatic Detecting Technology and Instruments under Grant YQ22108, and in part by the Innovation Project of GUET Graduate Education under Grant 2022YCXS161. We would like to thank the authors of [24,36,37] for making their code and data free for academic use.

    The authors declare no conflicts of interest.



    [1] E. J. Candes, J. Romberg, T. Tao, Robust uncertainty principles: exact signal reconstruction from highly incomplete frequency information, IEEE Trans. Inf. Theory, 52 (2006), 489–509. https://doi.org/10.1109/TIT.2005.862083 doi: 10.1109/TIT.2005.862083
    [2] E. T. Hale, W. Yin, Y. Zhang, Fixed-point continuation for l1-minimization: methodology and convergence, SIAM J. Optim., 19 (2008), 1107–1130. https://doi.org/10.1137/070698920 doi: 10.1137/070698920
    [3] Y. Li, C. Li, W. Yang, W. Zhang, A new conjugate gradient method with a restart direction and its application in image restoration, AIMS Math., 8 (2023), 28791–28807. https://doi.org/10.3934/math.20231475 doi: 10.3934/math.20231475
    [4] A. Chambolle, An algorithm for total variation minimization and applications, J. Math. Imag. Vis., 20 (2004), 89–97. https://doi.org/10.1023/B:JMIV.0000011325.36760.1e doi: 10.1023/B:JMIV.0000011325.36760.1e
    [5] L. Rudin, S. Osher, E. Fatemi, Nonlinear total variation based noise removal algorithms, Phys. D, 60 (1992), 259–268. https://doi.org/10.1016/0167-2789(92)90242-F doi: 10.1016/0167-2789(92)90242-F
    [6] N. Artsawang, Accelerated preconditioning Krasnosel'skii-Mann method for efficiently solving monotone inclusion problems, AIMS Math., 8 (2023), 28398–28412. https://doi.org/10.3934/math.20231453 doi: 10.3934/math.20231453
    [7] P. H. Calamai, J. J. Moré, Projeeted gradient methods for linearly constralned problems, Math. Program., 39 (1987), 93–116. https://doi.org/10.1007/BF02592073 doi: 10.1007/BF02592073
    [8] D. P. Bertsekas, Nonlinear programming, Athena scientific, Belmont, 1999.
    [9] J. Barzilai, J. M. Borwein, Two point step size gradient methods, IMA J. Numer. Anal., 8 (1988), 141–148. https://doi.org/10.1093/imanum/8.1.141 doi: 10.1093/imanum/8.1.141
    [10] E. G. Birgin, J. M. Martínez, M. Raydan, Nonmonotone spectral projected gradient methods on convex sets, SIAM J. Optim., 10 (2000), 1196–1211. https://doi.org/10.1137/S1052623497330963 doi: 10.1137/S1052623497330963
    [11] Y. H. Dai, R. Fletcher, Projected Barzilai-Borwein methods for large-scale box-constrained quadratic programming, Numer. Math., 100 (2005), 21–47. https://doi.org/10.1007/s00211-004-0569-y doi: 10.1007/s00211-004-0569-y
    [12] B. Zhang, Z. Zhu, S. Li, A modified spectral conjugate gradient projection algorithm for total variation image restoration, Appl. Math. Lett., 27 (2014), 26–35. https://doi.org/10.1016/j.aml.2013.08.006 doi: 10.1016/j.aml.2013.08.006
    [13] I. Daubechies, M. Defrise, C. De Mol, An iterative thresholding algorithm for linear inverse problems with a sparsity constraint, Commun. Pure Appl. Math., 57 (2004), 1413–1457. https://doi.org/10.1002/cpa.20042 doi: 10.1002/cpa.20042
    [14] A. Beck, M. Teboulle, A fast iterative shrinkage-thresholding algorithm for linear inverse problems, SIAM J. Imag. Sci., 2 (2009), 183–202. https://doi.org/10.1137/080716542 doi: 10.1137/080716542
    [15] E. T. Hale, W. Yin, Y. Zhang, Fixed-point continuation applied to compressed sensing: implementation and numerical experiments, J. Comput. Math., 28 (2010), 170–194. https://doi.org/10.4208/jcm.2009.10-m1007 doi: 10.4208/jcm.2009.10-m1007
    [16] S. J. Wright, R. D. Nowak, T. Figueiredo, Sparse reconstruction by separable approximation, IEEE Trans. Signal Process., 57 (2009), 2479–2493. https://doi.org/10.1109/TSP.2009.2016892 doi: 10.1109/TSP.2009.2016892
    [17] W. W. Hager, D. T. Phan, H. Zhang, Gradient-based methods for sparse recovery, SIAM J. Imag. Sci., 4 (2011), 146–165. https://doi.org/10.1137/090775063 doi: 10.1137/090775063
    [18] Y. Huang, H. Liu, A Barzilai-Borwein type method for minimizing composite functions, Numer. Algorithms, 69 (2015), 819–838. https://doi.org/10.1007/s11075-014-9927-8 doi: 10.1007/s11075-014-9927-8
    [19] W. Cheng, Z. Chen, D. Li, Nonmonotone spectral gradient method for sparse recovery, Inverse Probl. Imag., 9 (2015), 815–833. https://doi.org/10.3934/ipi.2015.9.815 doi: 10.3934/ipi.2015.9.815
    [20] Y. Xiao, S. Y. Wu, L. Qi, Nonmonotone Barzilai-Borwein gradient algorithm for l1-regularized nonsmooth minimization in compressive sensing, J. Sci. Comput., 61 (2014), 17–41. https://doi.org/10.1007/s10915-013-9815-8 doi: 10.1007/s10915-013-9815-8
    [21] P. Tseng, S. Yun, A coordinate gradient descent method for nonsmooth separable minimization, Math. Program., 117 (2009), 387–423. https://doi.org/10.1007/s10107-007-0170-0 doi: 10.1007/s10107-007-0170-0
    [22] E. Chouzenoux, J. C. Pesquet, A. Repetti, Variable metric forward-backward algorithm for minimizing the sum of a differentiable function and a convex function, J. Optim. Theory Appl., 162 (2014), 107–132. https://doi.org/10.1007/s10957-013-0465-7 doi: 10.1007/s10957-013-0465-7
    [23] P. L. Combettes, B. C. Vû, Variable metric forward-backward splitting with applications to monotone inclusions in duality, Optimization, 63 (2014), 1289–1318. https://doi.org/10.1080/02331934.2012.733883 doi: 10.1080/02331934.2012.733883
    [24] S. Bonettini, I. Loris, F. Porta, M. Prato, Variable metric inexact line-search-based methods for nonsmooth optimization, SIAM J. Optim., 26 (2016), 891–921. https://doi.org/10.1137/15M1019325 doi: 10.1137/15M1019325
    [25] S. Rebegoldi, L. Calatroni, Scaled, inexact and adaptive generalized FISTA for strongly convex optimization, SIAM J. Optim., 32 (2022), 2428–2459. https://doi.org/10.1137/21M1391699 doi: 10.1137/21M1391699
    [26] S.Bonettini, M. Prato, S. Rebegoldi, A new proximal heavy ball inexact line-search algorithm, Comput. Optim. Appl., 2024. https://doi.org/10.1007/s10589-024-00565-9 doi: 10.1007/s10589-024-00565-9
    [27] , S. Bonettini, P. Ochs, M. Prato, S. Rebegoldi, An abstract convergence framework with application to inertial inexact forward-backward methods, Comput. Optim. Appl., 84 (2023), 319–362. https://doi.org/10.1007/s10589-022-00441-4 doi: 10.1007/s10589-022-00441-4
    [28] H. Liu, T. Wang, Z. Liu, A nonmonotone accelerated proximal gradient method with variable stepsize strategy for nonsmooth and nonconvex minimization problems, J. Glob. Optim., 2024. https://doi.org/10.1007/s10898-024-01366-4 doi: 10.1007/s10898-024-01366-4
    [29] Y. Chen, W. W. Hager, M. Yashtini, X. Ye, H. Zhang, Bregman operator splitting with variable stepsize for total variation image reconstruction, Comput. Optim. Appl., 54 (2013), 317–342. https://doi.org/10.1007/s10589-012-9519-2 doi: 10.1007/s10589-012-9519-2
    [30] Y. Chen, W. W. Hager, F. Huang, D. T. Phan, X. Ye, W. Yin, Fast algorithms for image reconstruction with application to partially parallel MR imaging, SIAM J. Imag. Sci., 5 (2012), 90–118. https://doi.org/10.1137/100792688 doi: 10.1137/100792688
    [31] S. Bonettini, M. Prato, New convergence results for the scaled gradient projection method, Inverse Probl., 31 (2015), 095008. https://doi.org/10.1088/0266-5611/31/9/095008 doi: 10.1088/0266-5611/31/9/095008
    [32] F. Porta, M. Prato, L. Zanni, A new steplength selection for scaled gradient methods with application to image deblurring, J. Sci. Comput., 65 (2015), 895–919. https://doi.org/10.1007/s10915-015-9991-9 doi: 10.1007/s10915-015-9991-9
    [33] L. Grippo, F. Lampariello, S. Lucidi, A nonmonotone line search technique for Newton's method, SIAM J. Numer. Anal., 23 (1986), 707–716. https://doi.org/10.1137/0723046 doi: 10.1137/0723046
    [34] K. Amini, M. Ahookhosh, H. Nosratipour, An inexact line search approach using modified nonmonotone strategy for unconstrained optimization, Numer. Algorithms, 66 (2014), 49–78. https://doi.org/10.1007/s11075-013-9723-x doi: 10.1007/s11075-013-9723-x
    [35] R. T. Rockafellar, Convex analysis, Princeton University Press, 2015.
    [36] Y. Ouyang, Y. Chen, G. Lan, J. E. Pasiliao, An accelerated linearized alternating direction method of multipliers, SIAM J. Imag. Sci., 8 (2015), 644–681. https://doi.org/10.1137/14095697X doi: 10.1137/14095697X
    [37] X. Ye, Y. Chen, F. Huang, Computational acceleration for MR image reconstruction in partially parallel imaging, IEEE Trans. Med. Imag., 30 (2011), 1055–1063. https://doi.org/10.1109/TMI.2010.2073717 doi: 10.1109/TMI.2010.2073717
    [38] S. Bonettini, I. Loris, F. Porta, M. Prato, S. Rebegoldi, On the convergence of a linesearch based proximal-gradient method for nonconvex optimization, Inverse Probl., 33 (2017), 055005. https://doi.org/10.1088/1361-6420/aa5bfd doi: 10.1088/1361-6420/aa5bfd
    [39] S. Bonettini, M. Prato, S. Rebegoldi, Convergence of inexact forward-backward algorithms using the forward-backward envelope, SIAM J. Optim., 30 (2020), 3069–3097. https://doi.org/10.1137/19M1254155 doi: 10.1137/19M1254155
    [40] M. K. Riahi, I. A. Qattan, On the convergence rate of Fletcher-Reeves nonlinear conjugate gradient methods satisfying strong Wolfe conditions: application to parameter identification in problems governed by general dynamics, Math. Methods Appl. Sci., 45 (2022), 3644–3664. https://doi.org/10.1002/mma.8009 doi: 10.1002/mma.8009
    [41] X. Li, Q. L. Dong, A. Gibali, PMiCA-Parallel multi-step inertial contracting algorithm for solving common variational inclusions, J. Nonlinear Funct. Anal., 2022 (2022), 7. https://doi.org//10.23952/jnfa.2022.7 doi: 10.23952/jnfa.2022.7
    [42] L. O. Jolaoso, Y. Shehu, J. Yao, R. Xu, Double inertial parameters forward-backward splitting method: applications to compressed sensing, image processing, and SCAD penalty problems, J. Nonlinear Var. Anal., 7 (2023), 627–646. https://doi.org/10.23952/jnva.7.2023.4.10 doi: 10.23952/jnva.7.2023.4.10
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(842) PDF downloads(51) Cited by(0)

Other Articles By Authors

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog