JAMJournal of Applied Mathematics1687-00421110-757XHindawi Publishing Corporation24037010.1155/2011/240370240370Research ArticleA Class of Fourth-Order Telegraph-Diffusion Equations for Image RestorationZengWeili1LuXiaobo2TanXianghua3Van VleckE. S.1ITS Research CenterSchool of Transportation, Southeast UniversityNanjing 210096Chinaseu.edu.cn2School of AutomationSoutheast UniversityNanjing 210096Chinaseu.edu.cn3School of Mathematics and Computer ScienceHunan Normal UniversityNanjing 210096Chinahunnu.edu.cn20112606201120111403201117052011020620112011Copyright © 2011 Weili Zeng et al.This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

A class of nonlinear fourth-order telegraph-diffusion equations (TDE) for image restoration are proposed based on fourth-order TDE and bilateral filtering. The proposed model enjoys the benefits of both fourth-order TDE and bilateral filtering, which is not only edge preserving and robust to noise but also avoids the staircase effects. The existence, uniqueness, and stability of the solution for our model are proved. Experiment results show the effectiveness of the proposed model and demonstrate its superiority to the existing models.

1. Introduction

The use of second-order partial differential equations (PDE) has been studied as a useful tool for image restoration (noise removal). They include anisotropic diffusion equations  and total variation models  as well as curve evolution equations . These second-order techniques have been proved to be effective for removing noise without causing excessive smoothing of the edges. However, the images resulting from these techniques are often piecewise constant, and therefore, the processed image will look “blocky” . To reduce the blocky effect, while preserving sharp jump discontinuities, many methods in the literature  have been proposed to solve the problem.

A rather detailed analysis of blocky effects associated with anisotropic diffusion which was carried out in . Let u denote the image intensity function, t the time, the anisotropic diffusion equation as formulated by Perona and Malik  presented as (the PM model) ut-(g(|u|)u)=0, where g is the conductance coefficient, · and denote the divergence and the gradient, respectively. Equation (1.1) is associated with the following energy functional:E(u)=Ωf(|u|)dΩ, where Ω is the image support, and f(·)0 is an increasing function associated with the diffusion coefficient as g(s)=f(s)s. The PM model is then associated with an energy-dissipating process that seeks the minimum of the energy functional. From energy functional (1.2), it is obvious that level images are the global minima of the energy functional. Detailed analysis in  indicates that when there is no backward diffusion, a level image is the only minimum of the energy functional, so PM model will evolve toward the formation of a level image function. Since PM model is designed such that smooth areas are diffused faster than leee smooth ones, blocky effects will appear in the early stage of diffusion even though all the blocks will finally merge to form a level image. Similarly, when there is backward diffusion, however, any piecewise level image is a global minimum of the energy functional, so blocks will appear in the early stage of the diffusion and will.

In 2000, You and Kaveh  proposed the following fourth-order PDE (the YK model)ut+2(g(|2u|)2u)=0, where 2 denotes Laplace operator. The YK model replaces the gradient operator in PM model with a Laplace operator. Due to the fact that the Laplace of an image at a pixel is zero only if the image is planar in its neighborhood, the YK fourth-order PDE attempts to remove noise and preserve edges by approximating an observed image with a piecewise planar image. It is well known that piecewise smooth images look more natural than the piecewise constant images. Therefore, the blocky effect will be reduced and the image will look more nature. However, fourth-order PDE has an inherent limitation. The proposed PDE tends to leave images with speckle artifacts.

Recently, Ratner and Zeevi  introduced the following telegraph-diffusion equation (TDE model) utt+λut-(g(|u|)u)=0, where λ is the damping coefficient and g is the elasticity coefficient. Note that the TDE model is derived from (1.1) by adding second time derivative of the image. It is interesting to note that (1.5) converges to the diffusion equation (1.1) not only for large λ and g, but also after very long time [5, 6]. While one may find when the noise is large, (1.5) will be unstable in presence of noise which is similar to that of the PM model . In order to overcome this drawback, most recently, Cao et al.  proposed the following improved telegraph-diffusion equation (ITDE model)utt+λut-(g1(|Gσ*u|)u)=0, where Gσ is a Gaussian filtering. The TDE model (1.5), together with its improved version (1.6), has certain drawbacks. Since the model uses anisotropic diffusion, the filter inherits the staircase effects of anisotropic diffusion. The Gaussian filtering is not a perfect filtering method. Moreover, since the model uses anisotropic diffusion, the filter inherits the block effects of anisotropic diffusion.

Inspired by the ideas of [12, 22], it is natural to investigate a model inherits the advantages of the YK model and TDE model. Our proposed fourth-order telegraph-diffusion equation as follows: utt+λut+2(g2(|2u|)2u)=0, where 2 denotes Laplace operator. To our best knowledge, second-order derivative is sensitive to the noise. While one may find when the image is very noisy, (1.7) will be unstable which could not distinguish correctly the “true” edges and “false” edges. Considering that we can eliminate some noise before solving model (1.6), we reformulate model (1.6) as follows:utt+λut+2(g3(|2Bσ(u)|)2u)=0, where Bσ is a bilateral filtering [24, 25]; namely,Bσ(u(x))=1W(x)ΩGσs(ξ,x)Gσr(u(ξ),u(x))u(ξ)dξ, with the normalization constant W(x)=ΩGσs(ξ,x)Gσr(u(ξ),u(x))dξ, where Gσs will be a spatial Gaussian that decreases the influence of distant pixels, while Gσr will be a range Gaussian that decreases the influence of pixels ξ with intensity values that are very different from those of u(x); for example, Gσs=exp(-|ξ-x|22σs2),  Gσr=exp(-|u(ξ)-u(x)|22σr2), where parameters σs and σr  dictate the amount of filtering applied in the domain and the range of the image, respectively.

The ability of edge preservation in the fourth-order TDE-based image restoration method strongly depends on the conductance coefficient g. The desirable conductance coefficient should be diffused more in smooth areas and less around less intensity transitions, so that small variations in image intensity such as noise and unwanted texture are smoothed and edges are preserved. In the implementation of our proposed method, we use the following function: g3(s)=11+(s/k)2, where k is a threshold constant.

We will recall that the main purpose of the function g is to provide adaptive smoothing. It should not only precisely locate the position of the main edges, but it should also inhibit diffusion at edges. This is exactly what bilateral filtering accomplishes. It is proposed as a tool to reduce noise and preserve edges, by means of exploiting all relevant neighborhoods. It combines gray levels not only based on their gray similarity but also their geometric closeness and prefers near values to distant values in both domain and range.

The remainder of this paper is organized as follows. In Section 2, we will show the existence and uniqueness of our proposed model. Section 3 presents a discredited numerical implementation of the proposed model. Numerical experiments are presented in Section 4, and the paper is concluded in Section 5.

2. Analysis of Our Proposed Model: Existence and Uniqueness of Weak Solutions

In this section, we establish the existence and uniqueness of the following problem:utt+λut+2(g(|2Bσ(u)|)2u)=0,(x,t)ΩT=Ω×(0,T),u(x,0)=u0(x),ut(x,0)=0,xΩ,u=0,un=0,(x,t)Ω×(0,T), where Ω is a bounded domain of N with an appropriately smooth boundary, n denotes the unit outer normal to Ω, and T>0. In this section, C will represent a generic constant that may change from line to line even if in the same inequality.

The following standard notations are used throughout. We denote Hk(Ω) is a Hilbert space for the normuHk(Ω):=(|s|kΩ|muxm|2dx)1/2, where k is a positive integer and mu/xm of order |m|=j=1mmjk denotes the distributional derivative of u.

We denote by Lp(0,T;Hk(Ω)) the set of all functions u such that for almost every t in (0,T), u(t) belongs to Hk(Ω). Lp(0,T;Hk(Ω)) is a normed space for the norm uLp(0,T,Hk(Ω)):={(0Tu(t)Hk(Ω)p)1/p,(1p<),esssup0tTu(t)Hk(Ω),(p=), where k is a positive integer. L(0,T;L2(Ω)) is a normed space for the normuL(0,T,L2(Ω)):=esssup0tTu(t)L2(Ω). We denote by H2(Ω) the dual of H2(Ω) and introduce the following function space V(Ω)V(Ω)={uH2(Ω):u=0,  un,  xΩ}. Obviously, V(Ω) is a Banach space with the norm ||·||V(Ω)=||·||H2(Ω).

Definition 2.1.

Let T be a fixed positive number. A function u is a weak solution of the problem (2.1)–(2.3) provided

uC([0,T];  L2(Ω))L(0,  T;  V), utL(0,  T;  L2(Ω)) and uttL(0,  T;  V),

Ωuttφ+λΩutφ+Ωg(|ΔBσ*w|)ΔuΔφ=0 for any φV(Ω) and a. e. time 0tT,

u(0)=u0,  ut(0)=0.

2.1. Fourth-Order Linear Equation: Existence and Uniqueness

Now, we consider the existence and uniqueness of weak solutions of the following linear TD problemutt+λut+2(g(|2Bσ(w)|)2u)=0,(x,t)ΩT=Ω×(0,T),u(x,0)=u0(x),ut(x,0)=0,xΩ,u=0,un=0,(x,t)Ω×(0,T), where wL([0,T];L2(Ω)).

Definition 2.2.

A function uw is called a weak solution of the problem (2.8)–(2.10) provided

uwC([0,T];L2(Ω))L(0,T;V), (uw)tL(0,T;L2(Ω)) and (uw)ttL2(0,T;V),

Ω(uw)ttφdx+λΩ(uw)tφdx+Ωg(|2Bσ*w|)2uw2φdx=0for any φV(Ω) and a. e. 0tT,

u(0)=u0,ut(0)=0.

Theorem 2.3.

Suppose that wL([0,T];L2(Ω)) and u0H2(Ω), then the problem (2.8)–(2.10) admits a unique weak solution uw.

In order to prove Theorem 2.3, we will prove the following two lemmas.

Lemma 2.4.

Suppose that uL2(Ω), then there exists a constant C1>0, depending only on σ, |Ω| and N, such that 2Bσ(u)L(Ω)C1uL2(Ω). Moreover, there exists a constant C2>0, depending only on σ, |Ω|, n such that g(|2Bσ(u)|)-g(|2Bσ(ν)|)L(Ω)C2u-νL2(Ω), for each u,νL2(Ω).

Proof.

By the definition of Bσ in (1.8), we then calculate 2Bσ(u)L(Ω)=1W(x)Ω|exp(-|ξ-x|2σs2)||exp(-|u(ξ)-u(x)|2σr2)||u(ξ)|  dξC(|Ω|,σ,N)uL2(Ω)Ω|exp(-|ξ-x|2σs2)||exp(-|u(ξ)-u(x)|2σr2)|dξC1uL2(Ω). Moreover, since g(s) and Bσ are smooth, we have g(|2Bσ(u)|)-g(|2Bσ(ν)|)L(Ω)C|2Bσ(u)|2-|2Bσ(ν)|2L(Ω)C|2Bσ(u)|+|2Bσ(ν)|L(Ω)2Bσ(u-ν)L(Ω)=C(|Ω|,σ,N)(uL2(Ω)+νL2(Ω))u-νL2(Ω)C2(u-ν)L2(Ω). Therefore, we draw the conclusion.

Lemma 2.5.

If there exists M>0 such that wL(0,T;L2(Ω))+wtL(0,T;L2(Ω))M. Then, one has the estimate u(t)L(0,T;H2(Ω))+ut(t)L(0,T;L2(Ω))+uttL2(0,T;V)Cu0H2(Ω).

Proof.

Multiplying (2.8) by ut and integrating over Ω yields 12ddtΩ(ut)2dx+λΩ(ut)2dx+Ωg(|2Bσ(w)|)2u2utdx=0, for a.e. 0tT.

Note that Ωg(|2Bσ(w)|)ΔuΔutdx=ddt(12Ωg(|2Bσ(w)|)(2u)2dx)-12Ωgt(|2Bσ(w)|)(2u)2dx. Since w and wt satisfy (2.15), then 2Bσ(w), 2Bσ(wt) belong to L(0,  T;  L(Ω)) and there exists a constant C depending on Bσ and Ω such that 2Bσ(w)L(0,T;L(Ω)Cu0H2(Ω),2Bσ(wt)L(0,T;L(Ω)Cu0H2(Ω), for any xV, a.e. 0tT. Therefore, since g is decreasing and tg(t) is smooth, there exist two constants α, β>0 such that αg(|2Bσ(w)|)1,|gt(|2Bσ(w)|)|βu0H2(Ω), for any xΩ, a.e. 0tT.

Consequently, (2.18) yields Ωg(|2Bσ*w|)ΔuΔutdxα2ddt(Ω(2u)2dx)-CuH2(Ω)2. In terms of (2.17) and (2.21), it then follows that ddt(utL2(Ω)2+uH2(Ω)2)C(utL2(Ω)2+uH2(Ω)2), for a.e. 0tT.

Now, write η(t)=utL2(Ω)2+uH2(Ω)2. Then, (2.22) reads η(t)Cη(t), for a.e. 0tT. Applying the Gronwall inequality to (2.24) yields the estimate η(t)eCtη(0),0tT. Since η(0)=ut(0)L2(Ω)2+u(0)H2(Ω)2=u(0)H2(Ω)2, we obtain from (2.23)–(2.25) the estimate u(t)L(0,T;H2(Ω))+ut(t)L(0,T;L2(Ω))Cu0H2(Ω)2. Multiplying (2.8) by φ and integrating over Ω, we deduce for a.e. 0tT that Ωuttφdx+λΩutφdx+Ωg(|2Bσ*w|)2u2φdx=0, where φH02(Ω) and φH02(Ω)1.

Consequently, uttVCuH2(Ω). Therefore, 0TuttV2dtC0TuH2(Ω)2dtCu0H2(Ω)2. This completes the proof of the lemma.

Remark 2.6.

By Lemma 2.5, Theorem 2.3 can be proved by the Galerkin method (see ). Here, we omit the details of the proof of Theorem 2.3.

2.2. Fourth-Order Nonlinear Equation: Existence and UniquenessTheorem 2.7 (see ([<xref ref-type="bibr" rid="B26">26</xref>], Schauder’s Fixed Point Theorem)).

Suppose that X denotes a real Banach space and KX is compact and convex, and assume also that S:KX is continuous. Then, S has a fixed point in K.

Theorem 2.8.

Suppose that u0H2(Ω), then the problem (2.1)–(2.3) admits one and only one weak solution.

Proof.

In the following, we prove the theorem in two parts.(1) Existence of a Solution

In this first section, we show the existence of a weak solution of (2.1)–(2.3) by Shauder’s fixed point theorem  and Theorem 2.3. We introduce the space W(0,T)={wL(0,T;H2(Ω)),wtL(0,T;L2(Ω)),wttL(0,T;H2(Ω))}. Obviously, W(0,T) is a Banach space with the norm wW(0,T)=wL(0,T;H2(Ω))+wtL(0,T;L2(Ω))+wttL(0,T;H2(Ω)). Recalling Lemma 2.5, we introduce the subspace W0 of W(0,T) defined by W0={wW(0,T);w(0)=u0,  wt(0)=0,wL(0,T;L2(Ω))+wtL(0,T;L2(Ω))+wttL2(0,T;V)Cu0H2(Ω)}. By construction, S:wuw is a mapping from W0 into W0. Since W(0,T) is compactly imbedded in L2(0,T;L2(Ω)), W0 is a nonempty, convex, and weakly compact subset of L2(0,T;L2(Ω)).

In order to apply the Schauder fixed point theorem, we need to prove that S is a continuous compact mapping from W0 into W0. Let {wk} be a sequence in W0 which converges weakly to some w in W0 and uk=S(wk). We have to prove that S(wk)=uk converges weakly to S(w)=uw. By using the classical theorem of compact inclusion in Sobolev spaces , the sequence {wk} contains a subsequence (still denoted by) {wk} and {uk} contains a subsequence (still denoted by) {uk} such that wkwin  L2(0,T;L2(Ω))  and  a.e.  on  Ω×(0,T),g(|2Bσ*wk|)g(|2Bσ*w|)in  L2(0,T;L2(Ω))  and  a.e.  on  ΩT,ukuin  L2(0,T;H2(Ω)),(uk)t(u)tweakly  in  L2(0,T;L2(Ω)),(uk)tt(u)ttweakly  in  L2(0,T;V),2uk2uweakly  in  L2(0,T;L2(Ω)),uk(x,0)u(x,0)in  L2(Ω),(uk)t(x,0)0in  V. Since {uk} is the weak solutions of (2.8)–(2.10) corresponding to {wk}, we have, for a.e. 0tTΩ(uk)ttφdx+λΩ(uk)tφdx+Ωg(|2Bσ(wk)|)2uk2φdx=0. Integrating (2.43) from 0 to T yields 0TΩ(uk)ttφdxdt+λ0TΩ(uk)tφdxdt+0TΩg(|2Bσ(wk)|)2uk2φdxdt=0. From (2.38)-(2.39), we have 0TΩ(uk)ttφdxdt0TΩ(u)ttφdxdt,as  k,λ0TΩ(uk)tφdxdtλ0TΩ(u)tφdxdt. Considering the third term in (2.44) |0TΩg(|2Bσ(wk)|)2uk2φdxdt-0TΩg(|2Bσ(w)|)ΔuΔφdxdt|0TΩ|g(|2Bσ(wk)|)-g(|2Bσ(w)|)|ΔukΔφdxdt+|0TΩg(|2Bσ(wk)|)2(uk-u)2φdxdt|=A+B. According to Lemma 2.4 and (2.40), we deduce, as kAC0TΩwk-wL2(Ω)|Δuk||Δφ|C(φ)wk-wL2(Ω)ΔukL2(0,T;L2(Ω))0, and note also, as k, B.

Letting k in (2.44) yields 0TΩ(u)ttφdxdt+λ0TΩ(u)tφdxdt+0TΩg(|2Bσ(w)|)2u2φdxdt=0. Therefore, by the arbitrariness of φV, we obtain, for a.e. t[0,T]Ω(u)ttφdx+λΩ(u)tφdx+Ωg(|2Bσ(w)|)2u2φdx=0, which implies u=uw=S(w). By the uniqueness of the solution of linear problem (2.8)–(2.10) and (2.37), the sequence uk=S(wk) converges weakly in W0 to u=S(w). Hence, the mapping S is weakly continuous from W0 into W0. This in turn shows that mapping S is compact. A similar argument shows that S is a continuous mapping. Applying the Schauder fixed point theorem, we conclude that S has a fixed point u=S(u), which consequently solve (2.1)–(2.3). Using the classical theory of parabolic equations and the bootstrap argument , we can deduce that u is a strong solution of (1.1)–(1.4) and uC((0,T)×Ω).

(2) Uniqueness of the Solution

Now, we turn to the proof of the uniqueness, following the idea in . Let u1 and u2 be two solutions of (2.1)–(2.3), and we have, for almost every t in [0,T] and i=1,  2(u1-u2)tt+λ(u1-u2)t+2(β12(u1-u2))=2((β1-β2)2u2),(x,t)ΩT,ui(x,0)=u0(x),uit(x,0)=0,xΩ,ui=0,uin=0,(x,t)Ω×(0,T), in the distribution sense, where βi(t)=g(|2Bσ*ui|). It suffices to show that u1-u20. To verify this, fix 0<s<T and set νi(t)={tsui(τ)dτ,0<ts,0,st<T, for i=1,  2. Then, νi(t)H02(Ω) for each 0tT. Multiplying both sides of (2.52) by ν1-ν2, and integrating over Ω×(0,s) yields 0sΩ(-(u1-u2)t(ν1-ν2)t-λ(u1-u2)(ν1-ν2)t+β12(u1-u2)2(ν1-ν2))dxdt=0sΩ(β1-β2)2u22(ν1-ν2)dxdt. Now, (νi)t=-u (0<t<), and so 12Ω|(u1-u2)(,s)|2dx+0sΩλ|(u1-u2)|2dxdt+12Ωβ1(,0)|2(ν1-ν2)(,0)|2dx=0s(Ω(β1-β2)2u22(ν1-ν2)dx+12Ω|2(ν1-ν2)|2β1tdx)dt. As seen in the proof of Lemma 2.5, there exists positive constants C3 and C4 are positive constant depending on Bσ,  Ω,  Tand ||u0||H2(Ω) such that C3βi(,0)1,C3βi1,|(β1)t|C4,xΩ,  a.e.  t(0,T),  i=1,2, which imply from (2.58), 12Ω|(u1-u2)(,s)|2dx+0sΩλ|(u1-u2)|2  dxdt+C32Ω|2(ν1-ν2)(,0)|2dx0s(β1-β2L(Ω)(Ω|2u2|2dx)1/2(Ω|2(ν1-ν2)|2dx)1/2+C42Ω|2(ν1-ν2)|2dx)dt. By using the Young inequality to (2.60), we obtain 12Ω|(u1-u2)(,s)|2dx+0sΩλ|(u1-u2)|2  dxdt+C32Ω|2(ν1-ν2)(,0)|2dxC0s((Ω|u1-u2|2dx)1/2(Ω|2(ν1-ν2)|2dx)1/2+Ω|2(ν1-ν2)|2dx)dtC0s(Ω|u1-u2|2dx+Ω|2(ν1-ν2)|2dx)dt. Now, let us write wi(,t)=0tui(,τ)dτ,0<t<T, whereupon (2.61) can be rewritten as 12Ω|(u1-u2)(,s)|2dx+0sΩλ|(u1-u2)|2  dxdt+C32Ω|2(w1-w2)(,s)|2dxC0s(Ω|u1-u2|2dx+Ω|2(w1-w2)(,s)-2(w1-w2)(,t)|2dx)dt. Note that 2(w1-w2)(,s)-2(w1-w2)(,t)L2(Ω)222(w1-w2)(,s)L2(Ω)2+22(w1-w2)(,t)L2(Ω)2. Therefore, (2.63) implies 12Ω|(u1-u2)(,s)|2dx+0sΩλ|(u1-u2)|2  dxdt+C32Ω|2(w1-w2)(,s)|2dxC0s(Ω|u1-u2|2dx+2Ω|2(w1-w2)(,t)|2dx)dt+2Cs2(w1-w2)(,s)L2(Ω)2. Choose T1 so small such that C32-2CT1C34. Then, if 0<sT1, we have 12Ω|(u1-u2)(,s)|2dx+0sΩλ|(u1-u2)|2  dxdt+Ω|2(w1-w2)(,s)|2dxC0s(Ω|u1-u2|2dx+2Ω|2(w1-w2)(,t)|2dx)dt. Consequently, the integral form of the Gronwall inequality implies u1-u20 on (0,T1].

Finally, we apply the same argument on the intervals (T1,T2], (2T1,3T1], and so forth and eventually deduce that u1u2 on (0,T). Thus, we obtain the uniqueness of weak solutions.

3. Discretised Numerical Scheme

In this section, we construct an explicit discrete scheme to numerically solve differential equation (2.1)–(2.3). Assume a space grid size of h and a time step size of τ, we quantize the space and time coordinates as follows: t=n*τ,n=0,1,2,,x=i*h,i=0,1,2,,N,y=j*h,j=0,1,2,,M,

where M×N is the size of the image. Let ui,jn be the approximation to the value u(ih,jh,nτ). We then use a five-step approach to calculate (2.1)–(2.3).

(a) calculating the Laplace of the image intensity functions 2ui,jn=0.5[ui+1,jn+ui-1,jn+ui,j-1n+ui,j+1n-4ui,jnh2]+0.25[ui+1,j+1n+ui-1,j+1n+ui+1,j-1n+ui-1,j-1n-4ui,jnh2], with symmetric boundary conditions u-1,jn=u0,jn,uN+1,jn=uN,jn,j=0,1,2,,N,ui,-1n=ui,0n,ui,M+1n=ui,Mn,i=0,1,2,,M,

(b) calculating the value of the following functionφi,jn=c2(|2ui,jn|)2ui,jn,

(c) calculating the Laplace of φi,jn as2φi,jn=0.5[φi+1,jn+φi-1,jn+φi,j-1n+φi,j+1n-4φi,jnh2]+0.25[φi+1,j+1n+φi-1,j+1n+φi+1,j-1n+φi-1,j-1n-4φi,jnh2], with symmetry boundary conditionsφ-1,jn=φ0,jn,φN+1,jn=φN,jn,j=0,1,2,,N,φi,-1n=φi,0n,φi,M+1n=φi,Mn,i=0,1,2,,M,

(d) defining the discrete approximation:δui,jn=ui,jn-ui,jn-1τ,δ2ui,jn=δui,jn-δui,jn-1τ, with ui,j0=ui,j-1,i=0,1,2,,M,  j=0,1,2,,N,

(e) finally, the numerical approximation to the differential equation (2.1) is given as αδ2ui,jn+βδui,jn=-2φi,jn-1.

4. Experimental Results

In this section, we present numerical results obtained by applying our proposed fourth-order TDE to image denosing. We test the proposed method on “Barbara” image with size 225×225 (taken from USC-SIPI image database) and “license plate” image with size 240×306. These two images are shown in Figure 1(a) and Figure 3(a), respectively. The value chosen for the time step size τ is 0.25. To quantify the achieved performance improvements, we adopt improvement in signal to noise ratio (ISNR), which is defined asISNR=10log10(i,j[u(i,j)-u0(i,j)]2i,j[u(i,j)-unew(i,j)]2), where u0(·) is the initial image (noised image) and unew(·) is the denoised image. The value of ISNR is large, and the restored image is better.

Comparison of different methods on “Barbara” image. (a) Original image. (b) Noised image. (c) Fourth-order TDE with λ=1. (d) Fourth-order TDE with λ=40. (e) Fourth-order TDE with λ=70. (f) Fourth-order TDE with λ=100.

We first study the effects of damping coefficient λ. Figure 1(a) shows the noisy “Barbara” image. Figures 1(c)1(f) show the restored image using fourth-order TDE (k=0.5, a=4) with λ=1, 40, 70 and 100, respectively. In Figure 2, we plot the ISNR with different damping coefficient λ. We can see that the INSR reaches a maximum at λ=40. The INSR value using fourth-order TDE is lower than INSR value using proposed domain-based fourth-order PDE at λ<9.5. When λ100, the ISNR with fourth-order TDE is almost equal to the ISNR with domain-based fourth-order PDE.

Graph of ISNR versus different damping coefficient λ for “Barbara”.

Comparison of different methods on “License plate” image. (a) Original image. (b) Noisy image. (c) PM smodel with k=10. (d) TDE model with k=10, λ=40. (e) ITDE model with k=10, λ=40, σ1=0.1. (f) Proposed fourth-order TDE with k=1, a=4,λ=10.

Next, we test the proposed method for image restoration on 50 synthetic degraded images generated using random white Gaussian noise of variance σ=20. To verify the effectiveness of our proposed fourth-order TDE method for image restoration, it was evaluated in comparison with PM model , TDE model , and ITDE model . Figure 3(a) shows the original “License plate” image with size 240×306. We then added random white Gaussian noise of variance σ=20 to generate 50 degraded image. One such degraded image is shown in Figure 3(b). The results yield by PM second-order PDE is shown in Figure 3(c). We observe that “PM second-order PDE” can cause the processed image look block. The results yield by TDE and ITDE are shown in Figures 3(d) and 3(e), respectively. There two methods can leave much more sharp edges than PM second-order PDE but inherits the blocky effects from PM second-order PDE to some degree. Figure 3(f) is the restored image using our proposed fourth-order TDE. For the 50 random generated images, the mean of ISNR values for PM, TDE, ITDE, and our proposed fourth-order TDE are 5.7123, 5.8594, 5.9041, and 6.1688, respectively. Table 1 gives 10 of the 50 ISNR values. From Figure 3 and Table 1, we can conclude that our proposed method performs the best quality.

ISNR (in dB) values for “license plate” image using random white Gaussian noise of variance σ=20.

method12345678910
PM model5.72465.73095.74475.71275.75025.73545.74475.70495.72545.5188
TDE model5.83355.85365.84435.87515.82715.67475.83275.85735.87675.8187
ITDE model5.91475.91525.92175.91095.91985.91075.91225.90985.91505.9112
Fourth-order TDE6.17196.13826.16416.12686.16266.14316.15436.17976.11626.1883

Finally, we designed to further evaluate the good behavior of our proposed fourth-order TDE with white Gaussian noise across 5 noise levels. We added Gaussian white noise across five different variances σ to the original image. The ISNR values are given in Table 2. Our method obtains higher ISNR than the original method. Furthermore, a ISNR analysis conducted on the standard test image taken from USC-SIPI image database (http://sipi.usc.edu/database/) is listed in Table 3. We also note that our method obtains higher mean of ISNR than the original methods.

ISNR (in dB) values for “license plate” image using different methods across five noise levels.

 Method With Gaussian white noise σ=10 σ=15 σ=20 σ=25 σ=30 PM model 6.9930 6.5647 5.7246 4.9859 4.3677 TDE model 7.1032 6.6143 5.8335 5.1291 4.5038 ITDE model 7.2165 6.8372 5.9147 5.3022 4.6195 Fourth-order TDE 7.3648 6.9296 6.1719 5.5536 4.8564

The mean of ISNR (in dB) values for eight standard test images using random white Gaussian noise of variance 20.

 Methods Noise images Lena Boats Cameraman Peppers House Elaine PM model 4.3275 4.1487 4.0326 5.1120 4.2726 5.1737 TDE model 4.4662 4.2745 4.1689 5.2022 4.3996 5.2461 ITDE model 4.5743 4.3948 4.2917 5.2872 4.5127 5.3830 Fourth-order TDE 4.7125 4.5222 4.4134 5.4431 4.6475 5.5705
5. Conclusions

A class of nonlinear fourth-order telegraph-diffusion equation (TDE) for image restoration is presented in this paper. The proposed model first extends the second order TDE for image restoration to fourth-order TDE. Moreover, our proposed model combines nonlinear fourth-order TDE with bilateral filtering, which is not only edge preserving and robust to noise but also avoids the staircase effects. Finally, we study the existence, uniqueness, and stability of the proposed model. A set of numerical experiments is presented to show the good performance of our proposed model. Numerical results indicate that the proposed model recovers well edges and reduces noise. In [28, 29], the authors point out that the main disadvantage of higher order methods is the complexity of computation. In the future, we will explore a fast and efficient algorithm for our proposed fourth-order TDE method.

Acknowledgments

This work was supported by National Natural Science Foundation of China under Grant no. 60972001 and National Key Technologies R & D Program of China under Grant no. 2009BAG13A06. The authors would like to acknowledge Professor Qilin Liu and Yuxiang Li from Department of Mathematics for many fruitful discussions. The authors also thank the anonymous reviewer for his or her constructive and valuable comments, which helped in improving the presentation of our work.

PeronaP.MalikJ.Scale-space and edge detection using anisotropic diffusionIEEE Transactions on Pattern Analysis and Machine Intelligence19901276296392-s2.0-002546514510.1109/34.56205CattéF.LionsP.-L.MorelJ.-M.CollT.Image selective smoothing and edge detection by nonlinear diffusionSIAM Journal on Numerical Analysis1992291182193114909210.1137/0729012ZBL0746.65091WeiG. W.Generalized Perona-Malik equation for image restorationIEEE Signal Processing Letters1999671651672-s2.0-003262787110.1109/97.769359RudinL. I.OsherS.FatemiE.Nonlinear total variation based noise removal algorithmsPhysica D1992601–42592682-s2.0-44049111982KimiaB. B.TannenbaumA.ZuckerS. W.On the evolution of curves via a function of curvature. I. The classical caseJournal of Mathematical Analysis and Applications19921632438458114584010.1016/0022-247X(92)90260-KZBL0771.53003YouY. L.XuW.TannenbaumA.KavehM.Behavioral analysis of anisotropic diffusion in image processingIEEE Transactions on Image Processing1996511153915532-s2.0-0030283909JooK.KimS.PDE-based image restoration, I: anti-staircasing and anti-diffusion20032003-07Department of Mathematics, University of KentuckySavageJ.ChenK.On multigrids for solving a class of improved total variation based staircasing reduction modelsImage Processing Based on Partial Differential Equations: Mathematics and Visualization2007Berlin, GermanySpringer6994Math. Vis.10.1007/978-3-540-33267-1_52424222BaeE.Egil.Bae@math.uib.noShiJ.shij0004@e.ntu.edu.sgTaiX.-C.tai@mi.uib.noGraph cuts for curvature based image denoisingIEEE Transactions on Image Processing20112051199121010.1109/TIP.2010.2090533ChenQ.MontesinosP.SunQ. S.Shen XiaD.Ramp preserving Perona-Malik modelSignal Processing2010906196319752-s2.0-7684909848210.1016/j.sigpro.2009.12.015ShihY.ReiC.WangH.A novel PDE based image restoration: convection-diffusion equation for image denoisingJournal of Computational and Applied Mathematics20092312771779254974110.1016/j.cam.2009.05.001ZBL1169.94006YouY.-L.KavehM.Fourth-order partial differential equations for noise removalIEEE Transactions on Image Processing200091017231730180756610.1109/83.869184ZBL0962.94011LysakerM.LundervoldA.TaiX. C.Noise removal using fourth-order partial differential equation with applications to medical magnetic resonance images in space and timeIEEE Transactions on Image Processing20031212157915892-s2.0-454427811910.1109/TIP.2003.819229LiuQ.YaoZ.KeY.Entropy solutions for a fourth-order nonlinear degenerate problem for noise removalNonlinear Analysis: Theory, Methods & Applications20076761908191810.1016/j.na.2006.08.0162326039ZBL1119.35031GreerJ. B.BertozziA. L.H1 solutions of a class of fourth order nonlinear equations for image processingDiscrete and Continuous Dynamical Systems. Series A2004101-23493662026199ZBL1159.68619GreerJ. B.BertozziA. L.Traveling wave solutions of fourth order PDEs for image processingSIAM Journal on Mathematical Analysis2004361386810.1137/S00361410034273732083852ZBL1082.35080LiuQ.YaoZ.KeY.Solutions of fourth-order partial differential equations in a noise removal modelElectronic Journal of Differential Equations20071201112349948ZBL1138.35353DidasS.WeickertJ.BurgethB.Properties of higher order nonlinear diffusion filteringJournal of Mathematical Imaging and Vision2009353208226253393510.1007/s10851-009-0166-xLysakerM.TaiX. C.Iterative image restoration combining total variation minimization and a second-order functionalInternational Journal of Computer Vision20066615182-s2.0-2934444668610.1007/s11263-005-3219-7LiF.ShenC.FanJ.ShenC.Image restoration combining a total variational filter and a fourth-order filterJournal of Visual Communication and Image Representation20071843223302-s2.0-3425079683810.1016/j.jvcir.2007.04.005YiD.LeeS.Fourth-order partial differential equations for image enhancementApplied Mathematics and Computation20061751430440221635210.1016/j.amc.2005.07.043ZBL1124.65096RatnerV.ZeeviY. Y.Image enhancement using elastic manifoldsthe 14th Edition of the International Conference on Image Analysis and Processing (ICIAP '07)September 20077697742-s2.0-4814908911810.1109/ICIAP.2007.4362869CaoY.YinJ.LiuQ.LiM.A class of nonlinear parabolic-hyperbolic equations applied to image restorationNonlinear Analysis: Real World Applications201011125326110.1016/j.nonrwa.2008.11.0042570545ZBL1180.35378TomasiC.ManduchiR.Bilateral filtering for gray and color imagesProceedings of IEEE International Conference on Computer VisionJanuary 1998Bombay, India839846ZhangM.GunturkB. K.Multiresolution bilateral filtering for image denoisingIEEE Transactions on Image Processing2008171223242333251743510.1109/TIP.2008.2006658EvansL. C.Partial Differential Equations199819Providence, RI, USAAmerican Mathematical Societyxviii+662Graduate Studies in Mathematics1625845AdamsR. A.Sobolev Spaces19756London, UKAcademic Pressxviii+268Pure and Applied Mathematics0450957KomodakisN.ParagiosN.Beyond pairwise energies: efficient optimization for higher-order mrfsIEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPR '09)June 2009298529922-s2.0-7045019924110.1109/CVPRW.2009.5206846TaiX. C.HahnJ.ChungG. J.A fast algorithm for Euler's elastic model using augmented Lagrangian methodSIAM Journal on Imaging Sciences20114313344