1. bookAHEAD OF PRINT
Detalles de la revista
License
Formato
Revista
eISSN
2444-8656
Primera edición
01 Jan 2016
Calendario de la edición
2 veces al año
Idiomas
Inglés
Acceso abierto

The Practice System of Physics and Electronics Courses in Higher Vocational Colleges Based on Fractional Differential Equations

Publicado en línea: 15 Jul 2022
Volumen & Edición: AHEAD OF PRINT
Páginas: -
Recibido: 13 Feb 2022
Aceptado: 23 Apr 2022
Detalles de la revista
License
Formato
Revista
eISSN
2444-8656
Primera edición
01 Jan 2016
Calendario de la edición
2 veces al año
Idiomas
Inglés
Introduction

Electronic image denoising is an important research topic in image processing. It is also an important step in image preprocessing. Image noise is mainly divided into additive and multiplicative [1]. At present, the removal of additive Gaussian noise is relatively mature. For example, the total variation minimization method can remove noise and preserve the edges of the image well. Some scholars have given corresponding algorithms for removing multiplicative noise. The noise contained in the image is signal-dependent and follows a Poisson distribution. The photoelectron noise caused in the photoelectric conversion process of the image sensor has such a density distribution function [2]. Some scholars have studied the suppression of image Poisson noise by the variational method. They attributed the removal of Poisson noise to the evolution of a second-order partial differential equation to a stable state [3]. Some scholars have overcome this shortcoming with fourth-order partial differential equations. Some scholars use the alternate projection method to reduce the amount of calculation.

Recently, some new integro-differential equation methods have been applied in image denoising. Some scholars have proposed a new integral-differential equation. This equation does a good job of removing noise and decomposing images into cartoon and texture components. Some scholars have proposed partial differential equations with fractional time and spatial derivatives [4]. However, its essence is still integro-differential equations. Some scholars have proposed integro-differential equations with time integration. This equation has certain advantages over the traditional partial differential equation method in removing Gaussian noise. In this paper, a new multi-scale hierarchical image representation model based on Poisson noise is established based on the classical minimum total variation (TV) model. Then we introduce continuous-time variables to get new integro-differential equations with time integrals. The new integro-differential equation is an inverse scale-space method. The method evolves from the zero image to the original noisy image. The paper discusses some properties of integro-differential equations and obtains the energy decomposition theorem [5]. Numerical experiments show that the proposed algorithm is superior to the classical TV and fourth-order differential equation algorithms.

TV model based on Poisson noise

The classical TV model is obtained from minimizing the energy functional of equation (1). ETV(u)=Ω|u|dΩ+βΩ(fu)2dΩ {E_{TV}}\left( u \right) = \int_\Omega {\left| {\nabla u} \right|d\Omega + \beta \int_\Omega {{{\left( {f - u} \right)}^2}\,d\Omega } } where β is the regularization parameter. The first term in formula (1) is called the regularization term. The second item is called the loyalty item. The TV model can remove Gaussian noise well and preserve the edges of the image well [6]. Some scholars first analyze the extreme value problem of image denoising from the perspective of Bayesian estimation and then construct a variational model of Poisson distribution characteristics. Assume that the image additive noise follows a Poisson distribution: Pμ(n)=eμμnn!,n0 {P_\mu }\left( n \right) = {{{e^{ - \mu }}{\mu ^n}} \over {n!}},\,n \ge 0

Both mean and variance are μ. We observe image f and estimate original noise-free image μ. We analyze this problem using maximum a posteriori estimation [7]. i(u(xi)f(xi)logu(xi))+λΩ|u|dΩ \sum\limits_i {\left( {u\left( {{x_i}} \right) - f\left( {{x_i}} \right)\log \,u\left( {{x_i}} \right)} \right) + \lambda \int_\Omega {\left| {\nabla u} \right|d\Omega } }

We introduce a continuous variable x, so equation (7) can be regarded as the discretization of the following functional: E(u)=Ω(uflogu)dΩ+Ωλ|u|dΩ E\left( u \right) = \int_\Omega {\left( {u - f\,\log \,u} \right)d\Omega + } \int_\Omega {\lambda \left| {\nabla u} \right|d\Omega }

Functional (4) is defined in uBV and satisfies log uL1 (Ω). Its minimization solution is essentially different from the functional formula (1). The minimization solution of functional formula (4) cannot keep the average gray value of the image [8]. So we get the following properties:

Theorem 1

Any given λ has ΩudΩ=ΩfdΩuBV \int_\Omega {ud\Omega = } \int_\Omega {fd\Omega - {{\left\| u \right\|}_{BV}}} . where uBV=Ω|u|dΩ {\left\| u \right\|_{BV}} = \int_\Omega {\left| {\nabla u} \right|d\Omega } is the BV semi-norm. Prove that the Euler-Lagrange equation of equation (4) is represented as f+λudiv(u|u|)=u f + \lambda udiv\left( {{{\nabla u} \over {\left| {\nabla u} \right|}}} \right) = u .

Integrate both ends and obtain ΩfdΩ+λΩudiv(u|u|)dΩ=ΩfdΩλ(u|u|,u)=ΩfdΩλuBV=ΩudΩ \matrix{\int_\Omega {fd\Omega + \lambda \int_\Omega {udiv\left( {{{\nabla u} \over {\left| {\nabla u} \right|}}} \right)d\Omega } } \hfill\cr { = \int_\Omega {fd\Omega - \lambda \left( {{{\nabla u} \over {\left| {\nabla u} \right|}},\nabla u} \right)} } \hfill \cr { = \int_\Omega {fd\Omega - \lambda {{\left\| u \right\|}_{BV}} = \int_\Omega {ud\Omega } } } \hfill \cr } by Green's formula. The proof is complete.

New integro-differential equations

The minimal solution of functional (4) is the denoised image. We rewrite functional (4) into the following equation: E(u)=Ω|u|dΩ+λΩ(uflogu)dΩ E\left( u \right) = \int_\Omega {\left| {\nabla u} \right|d\Omega + \lambda } \int_\Omega {\left( {u - f\,\log \,u} \right)d\Omega }

We can use functional (5) to solve the minimization of uλ for very small λ. This equation contains only the main contours of the image. As λ gets bigger and bigger, uλ contains more and more image details [9]. If f is decomposed at the initial scale λ0, we have uλ0=argminυ=uflogu{uBV+λ0ΩυdΩ} {u_{{\lambda _0}}} = \mathop {\arg \min}\limits_{\upsilon = u - f\log u} \left\{ {{{\left\| u \right\|}_{BV}} + {\lambda _0}\int_\Omega {\upsilon d\Omega } } \right\}

Where the residual image is υλ0 = fuλ0. It still contains some image details. So υλ0 can still be decomposed at scale λ1: uλ1=argminυ=uuλ0logu{|u|BV+λ1ΩυdΩ} {u_{{\lambda _1}}} = \mathop {\arg \min}\limits_{\upsilon = u - {u_{{\lambda _0}}}\log u} \left\{ {{{\left| u \right|}_{BV}} + {\lambda _1}\int_\Omega {\upsilon d\Omega } } \right\}

Where the residual image is υλ1 = υλ0υλ1. If the fixed scale parameter λ is replaced by a series of varying scale parameters λ1 < λ2 < ⋯, then a new multi-scale hierarchical image representation is obtained. We solve the minimal solution of the following series of functionals: uλj=argminυ=uuλjlogu{|u|BV+λ1ΩυdΩ} {u_{{\lambda _j}}} = \mathop {\arg \min}\limits_{\upsilon = u - {u_{{\lambda _j}}}\log u} \left\{ {{{\left| u \right|}_{BV}} + {\lambda _1}\int_\Omega {\upsilon d\Omega } } \right\}

Where the residual image is υλj = υλj−1uλj. We can obtain new integro-differential equations from this relationship [10]. The article introduces successive time steps τ and re-solves uλj in units of τ. In this way, the following equation can be obtained. uλj=argminυ=τuuλj1logτu{|u|BV+λjΩυdΩ} {u_{{\lambda _j}}} = \mathop {\arg \min}\limits_{\upsilon = \tau u - {u_{{\lambda _{{j^{ - 1}}}}}}\log \tau u} \left\{ {{{\left| u \right|}_{BV}} + {\lambda _j}\int_\Omega {\upsilon d\Omega } } \right\}

Among them, the residual image is υλj = υλj−1τuλj, so after the N + 1 step, the following multi-scale layered image is obtained. We denote it as f = τuλ0 + υλ0 = τuλ0 + τuλ0 + τuλ1 + υλ1 = ⋯ = τuλ0 + τuλ1 + ⋯ + τuλN + υλn. which is f=k=0Nτuλk+υλN f = \sum\limits_{k = 0}^N {\tau {u_{{\lambda _k}}}} + {\upsilon _{{\lambda _N}}}

The N+1 step of multi-scale image representation is expressed as follows: uλN=argminυ=τuυλN1logτu{|u|BV+λNΩυdΩ} {u_{{\lambda _N}}} = \mathop {\arg \min}\limits_{\upsilon = \tau u - {\upsilon _{{\lambda _{N - 1}}}}\log \tau u} \left\{ {{{\left| u \right|}_{BV}} + {\lambda _N}\int_\Omega {\upsilon d\Omega } } \right\}

Where the residual image is υλN = υλN−1υλN. Its Euler-Lagrange equation is τuλNυλN1=1λNuλNdiv(uλN|uλN|) \tau {u_{{\lambda _N}}} - {\upsilon _{{\lambda _{N - 1}}}} = {1 \over {{\lambda _N}}}{u_{{\lambda _N}}}\,div\left( {{{\nabla {u_{{\lambda _N}}}} \over {\left| {\nabla {u_{{\lambda _N}}}} \right|}}} \right)

We have the expression υλN1=fk=0N1τuλk {\upsilon _{{\lambda _{N - 1}}}} = f - \sum\nolimits_{k = 0}^{N - 1} {\tau {u_{{\lambda _k}}}} from equation (10). We substitute it into equation (12) to get k=0Nτuλkf=1λNuλNdiv(uλN|uλN|) \sum\limits_{k = 0}^N {\tau {u_{{\lambda _k}}} - f = {1 \over {{\lambda _N}}}{u_{{\lambda _N}}}\,div\left( {{{\nabla {u_{{\lambda _N}}}} \over {\left| {\nabla {u_{{\lambda _N}}}} \right|}}} \right)}

Let τ → 0 get the new integro-differential equation (14): 0tu(x,y,s)dsf(x,y)=1λ(t)u(x,y,t)div(u(x,y,t)|u(x,y,t)|) \int_0^t {u\left( {x,\,y,\,s} \right)ds - f\left( {x,\,y} \right)} = {1 \over {\lambda \left( t \right)}}u\left( {x\,,y,\,t} \right)div\left( {{{\nabla u\left( {x\,,y,\,t} \right)} \over {\left| {\nabla u\left( {x\,,y,\,t} \right)} \right|}}} \right)

Where the initial condition is u(x, y, 0) = 0, and the boundary condition is un|Ω=0(t0) {{\partial u} \over {\partial n}}\left| {_{\partial \Omega } = 0\left( {t \ge 0} \right)} \right. . The scaling function λ(t) is monotonically increasing. In this paper, the image with Poisson noise is denoised by the integro-differential equation (14). The solution U(x,y,t)=0tu(x,y,s)ds U\left( {x,\,y,\,t} \right) = \int_0^t {u\left( {x,\,y,s} \right)} ds converges to the initial noisy image f when t → ∞. So the function family {U(x, y, t)}t≥0 can be regarded as the inverse scale-space representation of. t is the inverse scale parameter. Therefore we choose an appropriate time parameter t, U(x, y, t). V(x, y, t) = f(x, y) − U(x, y, t) is the residual image [11]. The new integro-differential equation has the following properties:

Theorem 2

We can obtain the following relation ΩUdΩ=ΩfdΩ1λ(t)uBV \int_\Omega {Ud\Omega = \int_\Omega {fd\Omega - {1 \over {\lambda \left( t \right)}}{{\left\| u \right\|}_{BV}}} } according to the integro-differential equation (14).

Theorem 3

We can obtain the energy decomposition formula: 40t1λ(s)Ωu(x,y,s)|u(x,y,s)|dΩds+V(x,y,t)L22=f|L22 4\int_0^t {{1 \over {\lambda \left( s \right)}}\int_\Omega {u\left( {x,\,y,\,s} \right)\left| {\nabla u\left( {x,\,y,\,s} \right)} \right|d\Omega ds + \left\| {V\left( {x,\,y,\,t} \right)} \right\|_{{L^2}}^2 = \left. {\left\| f \right\|} \right|_{{L^2}}^2} } according to the integro-differential equation (14). Ω(U(,t)f)ddt(U(,t)f)dΩ=12ddfΩ(U(,t)f)(U(,t)f)dΩ=Ω1λ(t)div(u(,t)|u(,t)|)u2(,t)dΩ \matrix{ {\int_\Omega {\left( {U\left( { \cdot ,t} \right) - f} \right){d \over {dt}}\left( {U\left( { \cdot ,t} \right) - f} \right)d\Omega } } \hfill \cr { = {1 \over 2}{d \over {df}}\int_\Omega {\left( {U\left( { \cdot ,t} \right) - f} \right)\,\left( {U\left( { \cdot ,t} \right) - f} \right)d\Omega } } \hfill \cr { = \int_\Omega {{1 \over {\lambda \left( t \right)}}div\left( {{{\nabla u\left( { \cdot ,t} \right)} \over {\left| {\nabla u\left( { \cdot ,t} \right)} \right|}}} \right){u^2}\,\left( { \cdot ,t} \right)d\Omega } } \hfill \cr }

Next, we perform time integration on both sides of equation (15) on [0, t] to get 12Ω(U(,t)f)(U(,t)f)dΩ12ΩffdΩstΩ1λ(s)div(u(,s)|u(,s)|)u2(,s)dΩds=U(,t)fL22fL22=s=0t2λ(s)(div(u(,s)|u(,s)|)u2(,s)ds) \matrix{ {{1 \over 2}\int_\Omega {\left( {U\left( { \cdot ,t} \right) - f} \right)\,\left( {U\left( { \cdot ,t} \right) - f} \right)d\Omega - {1 \over 2}\int_\Omega {f \cdot fd\Omega } } } \hfill \cr {\int_s^t {\int_\Omega {{1 \over {\lambda \left( s \right)}}div\left( {{{\nabla u\left( { \cdot ,s} \right)} \over {\left| {\nabla u\left( { \cdot ,s} \right)} \right|}}} \right){u^2}\,\left( { \cdot ,s} \right)d\Omega ds} } } \hfill \cr { = \left\| {U\left( { \cdot ,t} \right) - f} \right\|_{{L^2}}^2 - \left\| f \right\|_{{L^2}}^2} \hfill \cr { = \int_{s = 0}^t {{2 \over {\lambda \left( s \right)}}\left( {div\left( {{{\nabla u\left( { \cdot ,s} \right)} \over {\left| {\nabla u\left( { \cdot ,s} \right)} \right|}}} \right){u^2}\,\left( { \cdot ,s} \right)ds} \right)} } \hfill \cr }

We get the following equation from Green's formula: =s=0t2λ(s)(u(,s)|u(,s)|u2(,s))ds=s=0t2λ(s)Ω2u(,s)|u(,s)|dΩds \matrix{ { = - \int_{s = 0}^t {{2 \over {\lambda \left( s \right)}}\left( {{{\nabla u\left( { \cdot ,s} \right)} \over {\left| {\nabla u\left( { \cdot ,s} \right)} \right|}}{u^2}\,\left( { \cdot ,s} \right)} \right)ds} } \hfill \cr { = - \int_{s = 0}^t {{2 \over {\lambda \left( s \right)}}\int_\Omega {2u\,\left( { \cdot ,s} \right)\left| {\nabla u\left( { \cdot ,s} \right)} \right|d\Omega ds} } } \hfill \cr }

Our decomposition Theorem 3 gives the relationship between the residual images V and f.

There are further results if λ(t) satisfies a certain condition [12].

Theorem 4

Assume a noisy image fBV. From the integrodifferential equation (14) we can conclude that if the scaling function λ(t) increases fast enough to form a Hadamard sequence, limtλ(t/2)λ(t)=0 \mathop {\lim }\limits_{t \to \infty } {{\lambda \left( {t/2} \right)} \over {\lambda \left( t \right)}} = 0 , then there is Ωf(x,y)dΩ=Ω0u(x,y,s)dsdΩ \int_\Omega {f\left( {x,y} \right)d\Omega = \int_\Omega {\int_0^\infty {u\left( {x,y,s} \right)dsd\Omega } } } .

Proof

The residual image V(·, t) can be written as V(,t)=fs=0t/2u(,s)dss=t/2tu(,s)ds V\left( { \cdot ,t} \right) = f - \int_{s = 0}^{t/2} {u\left( { \cdot ,s} \right)ds - \int_{s = t/2}^t {u\left( { \cdot ,s} \right)ds} } . So there is inequality 1λ(t)V(,t)BV1λ(t)fBV+λ(t/2)λ(t)s=0t/21λ(s)u(,s)BV+s=t/2t1λ(s)u(,s)BVds \matrix{ {{1 \over {\lambda \left( t \right)}}{{\left\| {V\left( { \cdot ,t} \right)} \right\|}_{BV}} \le {1 \over {\lambda \left( t \right)}}{{\left\| f \right\|}_{BV}} + {{\lambda \left( {t/2} \right)} \over {\lambda \left( t \right)}}\int_{s = 0}^{t/2} {{1 \over {\lambda \left( s \right)}}} } \hfill \cr { \cdot {{\left\| {u\left( { \cdot ,s} \right)} \right\|}_{BV}} + \int_{s = t/2}^t {{1 \over {\lambda \left( s \right)}}{{\left\| {u\left( { \cdot ,s} \right)} \right\|}_{BV}}ds} } \hfill \cr }

Because when t → ∞ is λ(t)→∞, the first term on the right-hand side of the above inequality tends to 0. By hypothesis and theorem 3, it is found that the second and third terms also tend to 0. This way we get limt1λ(t)V(,t)BV=0 \mathop {\lim }\limits_{t \to \infty } {1 \over {\lambda \left( t \right)}}{\left\| {V\left( { \cdot ,t} \right)} \right\|_{BV}} = 0 . Because of ||u(·, t) ||BV ≤|| V(·, t)||BV, we have ΩV(,t)dΩ=1λ(t)u(,t)BV1λ(t)V(,t)BV \int_\Omega {V\left( { \cdot ,t} \right)d\Omega = {1 \over {\lambda \left( t \right)}}{{\left\| {u\left( { \cdot ,t} \right)} \right\|}_{BV}} \le {1 \over {\lambda \left( t \right)}}{{\left\| {V\left( { \cdot ,t} \right)} \right\|}_{BV}}} according to Theorem 2. So limtΩV(,t)dΩ=0 \mathop {\lim }\limits_{t \to \infty } \int_\Omega {V\left( { \cdot ,t} \right)d\Omega } = 0 .

Theorem 4 states that with the increase of time, the average gray value of the residual image converges to zero.

Numerical Algorithms and Experiments

We first assume that the initial discrete image f is of size m × m. It is the sampling of consecutive images at consistent sampling intervals [13]. The sampling point is (0, 0), ⋯, u(x, y) = u(xΔx, yΔy). where x, y = 0, ⋯, m − 1. We choose the spatial sampling interval Δx and Δy as Δx = Δy = 1. We make sure the denominator is not zero. We replace |∇u| with |u|2+ε \sqrt {{{\left| {\nabla u} \right|}^2} + \varepsilon } in equation (14). where ɛ is a small positive number. Jacobian iteration produces the sequence ωk+1Wn+1un+1 Δt when k → ∞. In the actual calculation, if ωk+1ωk220.01 \left\| {{\omega ^{k + 1}} - {\omega ^k}} \right\|_2^2 \le 0.01 , then stop the iteration. Next update image U: Un+1=Un+Wn+1 {U^{n + 1}} = {U^n} + {W^{n + 1}}

This paper selects the quadratic function λ(t) = 0.25 × t2 as the scaling function. Where the time step is taken as Δt = 1. Other types of monotonically increasing functions can also be used as scaling functions. According to numerical experiments, we found that the quadratic scaling function has fewer iterations and can achieve a good denoising effect. Figure 1 presents the evolution results at 4 scales [14]. The solution U of equation (15) contains only the blurred outline of the “Lenna” image. The solution U is already very close to the original noisy image f when t = 110. This shows that the new integro-differential equation is an inverse scale space method. As t and λ(t) increase, more and more details are added to the solution U.

Figure 1

Images at different scales using the integro-differential equation (22)

The second experiment added Poisson noise to the standard 512512′ images of “Lenna,” “Barbara”, “FishingBoat”, and “Pepper.” We use the signal-to-noise ratio (SNR) to measure the denoising effect and take Δt = 0.5 it as the time step. We choose an appropriate stop time t so that the signal-to-noise ratio reaches its maximum value. In Table 1, the signal-to-noise ratios of 3 different algorithms for denoising these 4 images. We can see that our algorithm achieves the highest signal-to-noise ratio [15]. Figure 2 shows the comparison of the denoising effects of the three algorithms on the “Lenna” image. We can see that the second-order differential equation method produces a block effect. The fourth-order differential equation method and the method in this paper have no block effect. However, our method preserves image details the best. This is mainly because the algorithm in this paper is an inverse scale space algorithm. The inverse scale space algorithm can better preserve the details of the image. Figure 3 shows the partial enlargement of the image after denoising by the three algorithms. We can see that the second-order differential equation algorithm has an obvious block effect. The image obtained by the algorithm in this paper has the best visual effect.

Comparison of signal-to-noise ratios of three algorithms (dB)

The original image Lenna Barbara fishing boat Pepper
Noisy image 4.13 4.18 4.71 4.38
Second order equation 11.65 8.86 10.78 13.05
Fourth order equation 11.84 10.07 10.81 13.11
Algorithm 13.35 10.44 11.14 13.65

Figure 2

Comparison of denoising effects of three algorithms

Figure 3

Comparison of local magnification of the three algorithms

Conclusion

We construct a new integro-differential equation denoising algorithm for Poisson-distributed noise in images. At the same time, we give a numerical solution method. The method in this paper has the following characteristics: (1) This paper firstly establishes a multi-scale hierarchical image representation model with a series of scale parameters. Then we serialize it to get new integro-differential equations with scaling functions. (2) The new integro-differential equation method is an inverse scale space method. It evolves the constant image to the original one, so an appropriate stopping time should be chosen to achieve a good denoising effect. Numerical experiments also prove that the proposed algorithm has a higher signal-to-noise ratio and visual effects than the classical TV and fourth-order differential equation algorithms.

Figure 1

Images at different scales using the integro-differential equation (22)
Images at different scales using the integro-differential equation (22)

Figure 2

Comparison of denoising effects of three algorithms
Comparison of denoising effects of three algorithms

Figure 3

Comparison of local magnification of the three algorithms
Comparison of local magnification of the three algorithms

Comparison of signal-to-noise ratios of three algorithms (dB)

The original image Lenna Barbara fishing boat Pepper
Noisy image 4.13 4.18 4.71 4.38
Second order equation 11.65 8.86 10.78 13.05
Fourth order equation 11.84 10.07 10.81 13.11
Algorithm 13.35 10.44 11.14 13.65

Jalab, H. A., Ibrahim, R. W., & Ahmed, A. Image denoising algorithm based on the convolution of fractional Tsallis entropy with the Riesz fractional derivative. Neural Computing and Applications., 2017; 28(1): 217–223 JalabH. A. IbrahimR. W. AhmedA. Image denoising algorithm based on the convolution of fractional Tsallis entropy with the Riesz fractional derivative Neural Computing and Applications 2017 28 1 217 223 10.1007/s00521-016-2331-7 Search in Google Scholar

Shukla, A. K., Pandey, R. K., Yadav, S., & Pachori, R. B. Generalized fractional filter-based algorithm for image denoising. Circuits, Systems, and Signal Processing., 2020; 39(1): 363–390 ShuklaA. K. PandeyR. K. YadavS. PachoriR. B. Generalized fractional filter-based algorithm for image denoising Circuits, Systems, and Signal Processing 2020 39 1 363 390 10.1007/s00034-019-01186-y Search in Google Scholar

Chen, D., Sun, S., Zhang, C., Chen, Y., & Xue, D. Fractional-order TV-L 2 model for image denoising. Central European Journal of Physics., 2013; 11(10): 1414–1422 ChenD. SunS. ZhangC. ChenY. XueD. Fractional-order TV-L 2 model for image denoising Central European Journal of Physics 2013 11 10 1414 1422 10.2478/s11534-013-0241-1 Search in Google Scholar

Huang, G., Xu, L., Chen, Q. L., & Pu, Y. F. Research on image denoising based on space fractional partial differential equations. Journal of Sichuan University: Engineering Science Edition., 2012; 44(2): 91–98 HuangG. XuL. ChenQ. L. PuY. F. Research on image denoising based on space fractional partial differential equations Journal of Sichuan University: Engineering Science Edition 2012 44 2 91 98 Search in Google Scholar

Jalab, H. A., & Ibrahim, R. W Fractional conway polynomials for image denoising with regularized fractional power parameters. Journal of Mathematical Imaging and Vision., 2015; 51(3): 442–450 JalabH. A. IbrahimR. W Fractional conway polynomials for image denoising with regularized fractional power parameters Journal of Mathematical Imaging and Vision 2015 51 3 442 450 10.1007/s10851-014-0534-z Search in Google Scholar

Zhou, P., Fan, Q. & Zhu, J. Empirical Analysis on Environmental Regulation Performance Measurement in Manufacturing Industry: A Case Study of Chongqing, China. Applied Mathematics and Nonlinear Sciences., 2020;5(1): 25–34 ZhouP. FanQ. ZhuJ. Empirical Analysis on Environmental Regulation Performance Measurement in Manufacturing Industry: A Case Study of Chongqing, China Applied Mathematics and Nonlinear Sciences 2020 5 1 25 34 10.2478/amns.2020.1.00003 Search in Google Scholar

Sulaiman, T., Bulut, H. & Baskonus, H. On the exact solutions to some system of complex nonlinear models. Applied Mathematics and Nonlinear Sciences., 2021; 6(1): 29–42 SulaimanT. BulutH. BaskonusH. On the exact solutions to some system of complex nonlinear models Applied Mathematics and Nonlinear Sciences 2021 6 1 29 42 10.2478/amns.2020.2.00007 Search in Google Scholar

Xu, L., Huang, G., Chen, Q. L., Qin, H. Y., Men, T., & Pu, Y. F. An improved method for image denoising based on fractional-order integration. Frontiers of Information Technology & Electronic Engineering., 2020;21(10): 1485–1493 XuL. HuangG. ChenQ. L. QinH. Y. MenT. PuY. F. An improved method for image denoising based on fractional-order integration Frontiers of Information Technology & Electronic Engineering 2020 21 10 1485 1493 10.1631/FITEE.1900727 Search in Google Scholar

Jalab, H. A., & Ibrahim, R. W. Fractional masks based on generalized fractional differential operator for image denoising. Int. J. Comput. Inf. Syst. Control Eng., 2013; 7(2): 169–174 JalabH. A. IbrahimR. W. Fractional masks based on generalized fractional differential operator for image denoising Int. J. Comput. Inf. Syst. Control Eng 2013 7 2 169 174 Search in Google Scholar

Ma, Q., Dong, F., & Kong, D. A fractional differential fidelity-based PDE model for image denoising. Machine Vision and Applications., 2017; 28(5): 635–647 MaQ. DongF. KongD. A fractional differential fidelity-based PDE model for image denoising Machine Vision and Applications 2017 28 5 635 647 10.1007/s00138-017-0857-z Search in Google Scholar

Jun, Z., & Zhihui, W. A class of fractional-order multi-scale variational models and alternating projection algorithm for image denoising. Applied Mathematical Modelling., 2011;35(5): 2516–2528 JunZ. ZhihuiW. A class of fractional-order multi-scale variational models and alternating projection algorithm for image denoising Applied Mathematical Modelling 2011 35 5 2516 2528 10.1016/j.apm.2010.11.049 Search in Google Scholar

Yu, J., Tan, L., Zhou, S., Wang, L., & Wang, C. Image denoising based on adaptive fractional order anisotropic diffusion. KSII Transactions on Internet and Information Systems (TIIS)., 2017;11(1): 436–450 YuJ. TanL. ZhouS. WangL. WangC. Image denoising based on adaptive fractional order anisotropic diffusion KSII Transactions on Internet and Information Systems (TIIS) 2017 11 1 436 450 Search in Google Scholar

Guidotti, P., & Longo, K. Two enhanced fourth order diffusion models for image denoising. Journal of Mathematical Imaging and Vision, 2011; 40(2): 188–198 GuidottiP. LongoK. Two enhanced fourth order diffusion models for image denoising Journal of Mathematical Imaging and Vision 2011 40 2 188 198 10.1007/s10851-010-0256-9 Search in Google Scholar

Jalab, H. A., & Ibrahim, R. W. Image denoising algorithms based on fractional sinc α with the covariance of fractional Gaussian fields. The Imaging Science Journal., 2016; 64(2): 100–108 JalabH. A. IbrahimR. W. Image denoising algorithms based on fractional sinc α with the covariance of fractional Gaussian fields The Imaging Science Journal 2016 64 2 100 108 10.1080/13682199.2016.1139522 Search in Google Scholar

Bai, J., & Feng, X. C. Fractional-order anisotropic diffusion for image denoising. IEEE transactions on image processing., 2007;16(10): 2492–2502 BaiJ. FengX. C. Fractional-order anisotropic diffusion for image denoising IEEE transactions on image processing 2007 16 10 2492 2502 10.1109/TIP.2007.904971 Search in Google Scholar

Artículos recomendados de Trend MD

Planifique su conferencia remota con Sciendo