CN105069762A - Image denoising method based on Shearlet transform and non-linear diffusion - Google Patents

Abstract

The invention discloses an image denoising method based on Shearlet transform and nonlinear diffusion, comprising steps: 1. Inputting the noisy image into an image processor; 2. The image processor performs multi-scale Shearlet transform on the noisy image to obtain low frequency Shearlet coefficients and high-frequency Shearlet coefficients, and the high-frequency Shearlet coefficients are recorded as initial noisy Shearlet coefficients; 3. The image processor continuously uses λ-time local Wiener filtering to perform diffusion and denoising processing on the initial noisy Shearlet coefficients; 4. Image The processor performs signal reconstruction to obtain a reconstructed image; 5. The image processor performs nonlinear diffusion post-processing on the reconstructed image to remove false details; 6. Outputs a denoised image. The method of the invention has simple steps, can effectively distinguish the signal from the noise, and protects the detailed features of the image while removing the noise, and has excellent performance.

Description

Image Denoising Method Based on Shearlet Transform and Nonlinear Diffusion

technical field

The invention belongs to the technical field of image processing, in particular to an image denoising method based on Shearlet transformation and nonlinear diffusion.

Background technique

Images are closely related to people's lives. However, images will be disturbed by various noises during the process of generation and transmission. Therefore, it is particularly important to denoise images and improve the subjective and objective effects of images. The methods and advantages and disadvantages of image denoising in the prior art are as follows:

On February 05, 2010, Liu Jinhua and others from the University of Electronic Science and Technology of China disclosed a dual-tree complex wavelet image denoising method based on partial differential equations in the Chinese patent application number 201010107623.0; on October 27, 2014, Liu Jinhua applied for The Chinese patent No. 201410584194.4 discloses an image denoising method based on dual-tree complex wavelet transform. This patent application discloses an improved image denoising method based on dual-tree complex wavelet transform and partial differential equation. The method implements a nonlinear diffusion scheme based on partial differential equations in the wavelet domain, which can remove noise well, but the dual-tree complex wavelet still lacks sufficient direction selectivity, which limits the room for performance improvement.

On July 23, 2014, Wang Guiting of Xidian University and others disclosed a two-dimensional Otsu-based contour wave domain Wiener filter image denoising method in the Chinese patent application number 201210061837.8. Carry out contourlet decomposition, and then perform two-dimensional Otsu segmentation on each decomposed high-frequency sub-band to obtain important coefficients and non-important coefficients; calculate the elliptical windows of high-frequency sub-bands respectively, and estimate the signal variance of high-frequency sub-bands according to the elliptical windows , Wiener filtering is performed on the important coefficients and non-important coefficients respectively; contourlet inverse transform is performed on the denoised high-frequency subbands to obtain the denoised image FI; non-local mean filtering is performed on FI to obtain the denoised output. This patent shows that its performance is better than the following three existing methods: one is better than the article "Image denoising via doubly Wiener filtering with adaptive directional windows and mean shift algorithm min wavelet domain" published by X.Li on pages 114-118 of "IEEE International Conference on Mechatronics and Automation (IEEE International Conference on Mechatronics and Automation)" in 2010. Adaptive window and mean shift algorithm wavelet domain double Wiener filter image denoising) "A wavelet domain double Wiener filter denoising method based on adaptive window is proposed; the second is better than Z.F.Zhou et al. in 2009 A contourlet based on an adaptive window is proposed in the article "Contourlet-based image denoising algorithm using adaptive windows (contourlet-based image denoising algorithm using adaptive window)" published on pages 3654-3657 of "IEEE Conference on Industrial Electronics and Applications". Domain Wiener filtering denoising method; the third is better than the article "Image denoising based on improved non-local means and non subsampled contour let transform Wiener filtering (based on improved Wiener filter image denoising based on non-local mean and non-subsampled contourlet transform)" proposed a non-subsampled contourlet domain Wiener filter denoising method based on non-local mean.

Although the "2D Otsu-based contour wave domain Wiener filter image denoising method" has achieved good results, but this method needs to classify wavelet coefficients, and after implementing Wiener filter denoising, non-local mean filtering is also required , which is complex to implement.

In order to solve the above problems, Miao Qiguang of Xidian University and others disclosed an image denoising method based on Shearlet transform and Wiener filter in the Chinese patent application number 201210364581.8. The method first symmetrically extends the input source image Then use the shear transformation, then use the wavelet packet decomposition, use the traditional Wiener filter for the decomposition coefficients, use the processed coefficients to obtain the reconstructed image, and finally perform the symmetric transformation and image fusion to obtain the final denoising image; this method uses Shearlet transform has the advantages of multi-directionality and Wiener filtering can adjust the filter output according to the regional variance of the image. The problem of unsatisfactory denoising effect caused by the same processing of coefficients in different directions, so that the image detail information can be more accurately analyzed in the high-frequency coefficients in different directions of the image, but there are false details in the generated image due to Wiener filtering , affecting the improvement of denoising performance.

Contents of the invention

The technical problem to be solved by the present invention is to provide an image denoising method based on Shearlet transform and nonlinear diffusion in view of the deficiencies in the above-mentioned prior art. The method has simple steps, is easy to implement, and can effectively distinguish signal from noise It can protect the detailed features of the image while removing the noise. It has excellent performance, strong practicability, good use effect, and is easy to promote and use.

For solving the problems of the technologies described above, the technical solution adopted in the present invention is: a kind of image denoising method based on Shearlet transformation and nonlinear diffusion, it is characterized in that the method comprises the following steps:

Step 1, inputting the noisy image into the image processor;

Step 2, the image processor performs multi-scale Shearlet transformation on the noisy image to obtain low-frequency Shearlet coefficients and high-frequency Shearlet coefficients, and record the high-frequency Shearlet coefficients as initial noisy Shearlet coefficients;

Step 3, the image processor continuously uses λ local Wiener filtering to perform diffusion and denoising processing on the initial noisy Shearlet coefficients; where λ is a natural number and the value of λ is 2 to 15;

Step 4, the image processor uses Shearlet inverse transform to perform signal reconstruction on the low-frequency Shearlet coefficients and the initial noisy Shearlet coefficients after the diffusion and denoising processing in step 3, to obtain a reconstructed image;

Step 5, the image processor performs nonlinear diffusion post-processing on the reconstructed image based on partial differential equations to remove false details of the reconstructed image;

Step 6, outputting the image after denoising.

The above-mentioned image denoising method based on Shearlet transform and nonlinear diffusion is characterized in that: in step 2, the image processor performs five-layer multi-scale Shearlet transform on the noisy image, and the first layer and the second layer two-stage high-frequency sub-band Each has 16 directions, the third and fourth layers of high-frequency sub-bands each have 8 directions, and the fifth layer is a low-frequency sub-band.

The above-mentioned image denoising method based on Shearlet transform and nonlinear diffusion is characterized in that: in step 3, the image processor continuously uses λ local Wiener filtering to carry out the specific process of diffusion and denoising processing on the initial noisy Shearlet coefficients as follows: The initial noisy Shearlet coefficient is used as the input signal of the first Wiener filter, and the output signal of the current Wiener filter is used as the input signal of the next Wiener filter; the rth Wiener filter is at the subband position (x, y) The recovered high-frequency Shearlet coefficients are denoted as z _r (x, y) and z _r (x, y) = W _r (x, y) z _r-1 (x, y), where the value of r is 1 ～λ is a natural number, z _r-1 (x, y) is the high-frequency Shearlet coefficient restored at the subband position (x, y) by the r-1 Wiener filter, z ₀ (x, y) is The initial noise coefficient at the subband position (x,y), W _r (x,y) is the rth Wiener filter and n _r (x,y) is the noise component to be removed at the subband position (x,y), is the variance of n _r (x,y) and takes η is a noise variance adjustment parameter and η is a real number, is the initial noise variance in the subband and It is estimated by the Monte-Carlo method; s _r (x, y) is the signal component to be recovered at the subband position (x, y), is the variance of s _r (x,y) and According to the formula $\mathrm{E.}\left\{{\mathrm{the\; s}}_r^2(x,\mathrm{the\; y})\right\}=\mathrm{E.}\left\{z_{r-1}^2(x,\mathrm{the\; y})\right\}-\mathrm{\δ}\mathrm{E.}\left\{{\mathrm{no}}_r^2(x,\mathrm{the\; y})\right\}=\frac{1}{{(2R+1)}^2}\underset{l1=-R}{\overset{R}{\Σ}}\underset{l2=-R}{\overset{R}{\Σ}}{z}_{r-1}^2(x-l1,\mathrm{the\; y}-l2)-\mathrm{\δ}\mathrm{\η}{\widehat{\mathrm{\σ}}}^2/\mathrm{\λ}$ It is estimated that is the variance of z _r-1 (x, y), δ is the adjustment parameter of the variance of s _r (x, y), and δ is a non-negative real number, R is a constant and a positive integer, l1 and l2 are both integer variables And the value is -R～R, z _r-1 (x-l1, y-l2) is the high frequency Shearlet coefficient, z ₀ (x-l1, y-l2) is the initial noise coefficient at subband position (x-l1, y-l2).

The above-mentioned image denoising method based on Shearlet transform and nonlinear diffusion is characterized in that the value of η is a real number ranging from 1 to 1.5.

The above-mentioned image denoising method based on Shearlet transform and nonlinear diffusion is characterized in that: in step five, the image processor performs nonlinear diffusion post-processing on the reconstructed image based on partial differential equations, and removes the details of false details of the reconstructed image. The process is: Assuming the reconstructed image is I, use partial differential equations for I Perform nonlinear diffusion post-processing, where div is the divergence operator, t is the evolution time of partial differential diffusion, and the value of t increases with the deepening of diffusion; I _t (i, j) is the position at time t (i, The gray value of the pixel at j), is the gradient of I _t (i,j), c _t (i,j) is the exponential diffusion function and PDE discretized as: $I_k\left(i,j\right)=I_{k-1}\left(i,j\right)+0.2\&Center\; Dot;\underset{u=1}{\overset{4}{\Σ}}\left(I_{k-1}^u\left(i,j\right)-I_{k-1}\left(i,j\right)\right)\exp \[-{\left(ab\mathrm{the\; s}\left(I_{k-1}^u\left(i,j\right)-I_{k-1}\left(i,j\right)\right)/K\right)}^2\],$ in, is the gray value of the pixel in the north, south, east, and west directions of (i, j), K is the gradient threshold and the value of K is 1 to 10, I _k (i, j) is the position in ( The pixel gray value after the k-th diffusion denoising at i, j), I _k-1 (i, j) is the pixel gray value after the k-1 diffusion de-noising at the position (i, j) I ₀ (i, j) is the pixel gray value of the reconstructed image I at (i, j), k is the number of iterations and the value of k is a positive integer between 1 and 7.

Compared with the prior art, the present invention has the following advantages:

1, the method step of the present invention is simple, realizes conveniently.

2. Compared with the prior art, the present invention can effectively distinguish the signal from the noise, and protect the details of the image while removing the noise, so as to obtain the best restoration of the original image.

3, the selected Shearlet of the present invention has sufficient direction selectivity, can well capture the geometric feature of image; The nonlinear diffusion based on the partial differential equation is implemented on the final image to remove the false details generated by the reconstructed image, and the performance is excellent.

4. The present invention has strong practicability, good use effect, and is convenient for popularization and use.

In summary, the method of the present invention has simple steps, is easy to implement, can effectively distinguish signal from noise, and protects the details of the image while removing noise, has excellent performance, strong practicability, and good use effect , which is convenient for promotion and use.

The technical solutions of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments.

Description of drawings

Fig. 1 is a flow chart of the method of the present invention.

Figure 2A is the original Barbara image.

Figure 2B is a noisy Barbara image.

Figure 2C is the denoised Barbara image processed by the WWF method.

Figure 2D is the denoised Barbara image processed by the GWF method.

Figure 2E is the denoised Barbara image processed by the LBS method.

Figure 2F is the denoised Barbara image processed by the ProbShrink method.

Figure 2G is the denoised Barbara image processed by the UWTSURE-LET method.

Fig. 2H is a denoised Barbara image processed by the method of the present invention.

Detailed ways

As shown in Figure 1, the image denoising method based on Shearlet transform and nonlinear diffusion of the present invention is characterized in that the method comprises the following steps:

Step 1, inputting the noisy image into the image processor;

Step 2, the image processor performs multi-scale Shearlet transformation on the noisy image to obtain low-frequency Shearlet coefficients and high-frequency Shearlet coefficients, and record the high-frequency Shearlet coefficients as initial noisy Shearlet coefficients;

Step 3, the image processor continuously uses λ local Wiener filtering to perform diffusion and denoising processing on the initial noisy Shearlet coefficients; where λ is a natural number and the value of λ is 2 to 15;

Step 4, the image processor uses Shearlet inverse transform to perform signal reconstruction on the low-frequency Shearlet coefficients and the initial noisy Shearlet coefficients after the diffusion and denoising processing in step 3, to obtain a reconstructed image;

Step 5, the image processor performs nonlinear diffusion post-processing on the reconstructed image based on partial differential equations to remove false details of the reconstructed image;

Step 6, outputting the image after denoising.

In this embodiment, in step 2, the image processor performs five-layer multi-scale Shearlet transformation on the noisy image. The first and second layers of high-frequency subbands each have 16 directions, and the third and fourth layers have two Each of the high-frequency sub-bands has 8 directions, and the fifth layer is the low-frequency sub-band.

During specific implementation, the Shearlet transformation adopted is the article "Sparse directional image representations using the discrete shearlet transform ( The discrete non-subsampling Shearlet transform proposed in "Oriented Sparse Representation of Image Based on Discrete Shearlet Transformation)" uses the window function "Meyer". The discrete Shearlet transform has translation invariance and good directionality, and can more effectively capture The geometry of the image.

In this embodiment, in step 3, the image processor continuously uses λ times of local Wiener filtering to perform diffusion and denoising processing on the initial noisy Shearlet coefficients: take the initial noisy Shearlet coefficients as the input of the first Wiener filtering signal, the output signal of the current Wiener filter is used as the input signal of the next Wiener filter; the high-frequency Shearlet coefficient recovered by the rth Wiener filter at the subband position (x, y) is denoted as z _r (x, y) and z _r (x, y) = W _r (x, y) z _r-1 (x, y), wherein, the value of r is a natural number from 1 to λ, z _r-1 (x, y) is the high-frequency Shearlet coefficient restored at the subband position (x, y) by the r-1 Wiener filter, and z ₀ (x, y) is the initial noise at the subband position (x, y) coefficient, W _r (x, y) is the rth Wiener filter and n _r (x,y) is the noise component to be removed at the subband position (x,y), is the variance of n _r (x,y) and takes η is a noise variance adjustment parameter and η is a real number, is the initial noise variance in the subband and It is estimated by the Monte-Carlo method; s _r (x, y) is the signal component to be recovered at the subband position (x, y), is the variance of s _r (x,y) and According to the formula $\mathrm{E.}\left\{{\mathrm{the\; s}}_r^2(x,\mathrm{the\; y})\right\}=\mathrm{E.}\left\{z_{r-1}^2(x,\mathrm{the\; y})\right\}-\mathrm{\δ}\mathrm{E.}\left\{{\mathrm{no}}_r^2(x,\mathrm{the\; y})\right\}=\frac{1}{{(2R+1)}^2}\underset{l1=-R}{\overset{R}{\Σ}}\underset{l2=-R}{\overset{R}{\Σ}}{z}_{r-1}^2(x-l1,\mathrm{the\; y}-l2)-\mathrm{\δ}\mathrm{\η}{\widehat{\mathrm{\σ}}}^2/\mathrm{\λ}$ It is estimated that is the variance of z _r-1 (x, y), δ is the adjustment parameter of the variance of s _r (x, y), and δ is a non-negative real number, R is a constant and a positive integer, l1 and l2 are both integer variables And the value is -R～R, z _r-1 (x-l1, y-l2) is the high frequency Shearlet coefficient, z ₀ (x-l1, y-l2) is the initial noise coefficient at subband position (x-l1, y-l2).

When the image processor performs five-layer multi-scale Shearlet transformation on the noisy image in step 2, when processing the Shearlet coefficients in the first layer of high-frequency sub-bands to the fourth layer of high-frequency sub-bands in step 3, the value of R The values are 7, 2, 2, 2 in sequence.

In this embodiment, the value of λ is 10, and the value of η is a real number ranging from 1 to 1.5. Preferably, the value of n is 1.2, namely That is, the noise variance of each Wiener filter in the present invention is the initial noise variance in the subband 1.2 times of 10, the signal variance of each Wiener filter is estimated by the current input signal, and take δ=0, that is, the present invention is to traditional Wiener filter The estimation method of is modified, after the modification, the variance value is less than The signal details of are no longer removed in the diffusion denoising process.

In this embodiment, in step 5, the image processor performs nonlinear diffusion post-processing on the reconstructed image based on partial differential equations, and the specific process of removing false details of the reconstructed image is as follows: Assuming that the reconstructed image is I, use partial differential equations for I differential equation Perform nonlinear diffusion post-processing, where div is the divergence operator, t is the evolution time of partial differential diffusion, and the value of t increases with the deepening of diffusion; I _t (i, j) is the position at time t (i, The gray value of the pixel at j), is the gradient of I _t (i,j), c _t (i,j) is the exponential diffusion function and PDE discretized as: $I_k\left(i,j\right)=I_{k-1}\left(i,j\right)+0.2\·\underset{u=1}{\overset{4}{\Σ}}\left(I_{k-1}^u\left(i,j\right)-I_{k-1}\left(i,j\right)\right)\exp \[-{\left(ab\mathrm{the\; s}\left(I_{k-1}^u\left(i,j\right)-I_{k-1}\left(i,j\right)\right)/K\right)}^2\],$ in, is the gray value of the pixel in the north, south, east, and west directions of (i, j), K is the gradient threshold and the value of K is 1 to 10, preferably 6, I _k (i, j) is the pixel gray value after the k-th diffusion denoising at the position (i,j), I _k-1 (i,j) is the k-1th diffusion denoising at the position (i,j) After the pixel gray value, I ₀ (i, j) is the pixel gray value of the reconstructed image I at (i, j), k is the number of iterations and the value of k is between 1 and 7 positive integer.

In this embodiment, the image processor is a computer.

In order to verify the technical effects that the present invention can produce, MATLAB7.0 software is used to carry out the following simulation demonstration; in the simulation process, the noise added to the test image is Gaussian white noise with a mean value of zero and different standard deviations σ.

Simulation 1

Adopt the Lena and Barbara image that size is 512 * 512 pixels as the test image, SD represents the method of the present invention, simulation 1 has provided the method of the present invention and the classical good wavelet domain image denoising algorithm and local Wiener filter that appear in the world The comparison results of image denoising algorithms, and the comparison of peak signal-to-noise ratio (PSNR) output by different denoising methods are shown in Table 1:

In Table 1, WWF is the article "Low complexity image denoising based on statistical modeling of wavelet coefficients" published by M.K.Mihcak et al. in "IEEE Signal Processing Letters (IEEE Signal Processing Letters)" Vol. Noise) is the classic "wavelet domain Wiener filter" image denoising method; GWF is X. Zhang et al. in 2013 in "Computers and Electrical Engineering (Computer and Electrical Engineering)", Volume 39, Issue 3, Page 934-344 The "Gradient-based Wiener filter for image denoising" proposed in the published article "Gradient-based Wiener filter for image denoising" (Gradient-based Wiener filter for image denoising) Express) "Volume 9, No. 12, Page 438-441, the article "Bivariateshrinkagewithlocal varianceestimation (bivariate shrinkage with local variance estimation)" proposed "bivariate shrinkage based on local variance estimation" image denoising method; ProbShrink is A.Pizurica et al. published the article "Estimating the probability of the presence of a signal of interest in multiresolution single-and multiband image denoising" in "IEEE Transactions on Image Processing (IEEE Image Processing Transactions)" Volume 15, No. 3, Pages 654-665 in 2006. Probability Estimation of the Existence of Signals)" proposes an image denoising method based on "shrinking based on the probability that wavelet coefficients contain important information"; UWTSURE-LET is T. The image denoising method "based on Stein unbiased risk estimation and threshold linear expansion" proposed in the article "TheSURE-LET approach to image denoising (image denoising based on SURE-LET method)" published on pages 2778-2786 of volume 16, issue 11 .

Table 1 Comparison of peak signal-to-noise ratio PSNR output by different denoising methods

It can be seen from Table 1 that the present invention is superior to other image denoising methods mentioned above, and has the highest peak signal-to-noise ratio (PSNR).

Figures 2A-2H show the comparison of visual effects of several methods for Barbara image processing with standard deviation σ=25. Among them, Figure 2A is the original Barbara image, Figure 2B is the noisy Barbara image, Figure 2C is the denoised Barbara image processed by the WWF method, Figure 2D is the denoised Barbara image processed by the GWF method, and Figure 2E is the denoised Barbara image processed by the LBS method. The denoising Barbara image after processing by the method, Fig. 2F is the denoising Barbara image after processing with the ProbShrink method, Fig. 2G is the denoising Barbara image after processing with the UWTSURE-LET method, Fig. 2H is the denoising Barbara image after processing with the method of the present invention Denoising the Barbara image.

It can be seen from Fig. 2A～2H that there are obvious false details in the denoised Barbara image processed by WWF method; there is a lot of noise in the denoised Barbara image processed by GWF method; although LBS method, ProbShrink method and The denoising Barbara image processed by the UWTSURE-LET method surpasses the denoising Barbara image processed by the WWF method and the GWF method in visual effect, but compared with the denoising Barbara image processed by the method of the present invention, there are still more fake details. The invention can better preserve the feature of the image while removing the noise, and has the best visual effect.

It can be seen from the above comparison that the present invention is superior to some other classic wavelet domain image denoising methods and local Wiener filtering methods in the world in both subjective and objective effects.

Simulation 2

The Lena and Barbara images with a size of 512×512 pixels are used as test images, SD represents the method of the present invention, and simulation 2 shows the comparison results between the method of the present invention and the good image denoising method based on the wavelet domain diffusion format in the world , the comparison of peak signal-to-noise ratio (PSNR) output by different denoising methods is shown in Table 2 and Table 3:

WMSAD is the article "Wavelet-basedmultiscaleanisotropicgiffusionwithadaptivestatisticalanalysisforimagerestoration" published by J.zhong et al. Multiscale Anisotropic Diffusion Image Restoration with Adaptive Statistical Analysis)" proposes a "nonlinear diffusion based on partial differential equation in wavelet domain" image denoising method; SWCD is A.K.Mandava et al. (Journal of Electronic Imaging) "Volume 20, Issue 3, Page 033016-1-033016-7 published the article "Image denoising based on adaptive non-linear diffusion in wavelet domain (wavelet domain adaptive non-linear diffusion image denoising)" proposed a "stationary wavelet domain using context Nonlinear diffusion of information" image denoising method.

Table 2 Comparison of output PSNR of SD and WMSAD

Table 3 Comparison of output PSNR of SD and SWCD

It can be seen from Table 2 and Table 3 that the present invention is superior to the existing image denoising method based on the diffusion scheme in the wavelet domain which is the best in the world, and has the highest peak signal-to-noise ratio (PSNR).

Simulation 3

The Lena and Barbara images with a size of 512×512 pixels are used as test images, SD represents the method of the present invention, simulation 3 shows the comparison results of the method of the present invention and technology 1 and technology 2, and the peak signal-to-noise ratio output by different denoising methods The comparison of (PSNR) is shown in Table 4 and Table 5:

Table 4 Comparison of output PSNR of SD and technique 1

Table 5 Comparison of output PSNR of SD and Technique 2

Technology 1 is an image denoising method based on Shearlet transform and Wiener filtering disclosed in the Chinese patent application number 201210364581.8 mentioned in the background technology, and technology 2 is disclosed in the Chinese patent application number 201210061837.8 mentioned in the background technology Image denoising method based on two-dimensional Otsu-based Wiener filter in contour wave domain.

It can be seen from Table 4 and Table 5 that the present invention is superior to the image denoising methods in Technology 1 and Technology 2, and has the highest peak signal-to-noise ratio (PSNR).

The above are only preferred embodiments of the present invention, and do not limit the present invention in any way. All simple modifications, changes and equivalent structural changes made to the above embodiments according to the technical essence of the present invention still belong to the technical aspects of the present invention. within the scope of protection of the scheme.

Claims (5)

1. An image denoising method based on Shearlet transformation and nonlinear diffusion is characterized by comprising the following steps:

step one, inputting a noise-containing image into an image processor;

step two, the image processor carries out multi-scale Shearlet transformation on the noisy image to obtain a low-frequency Shearlet coefficient and a high-frequency Shearlet coefficient, and the high-frequency Shearlet coefficient is marked as an initial noisy Shearlet coefficient;

step three, continuously using lambda times of local wiener filtering by the image processor to perform diffusion denoising processing on the initial noisy Shearlet coefficient; wherein lambda is a natural number and the value of lambda is 2-15;

fourthly, the image processor performs signal reconstruction on the low-frequency Shearlet coefficient and the initial noisy Shearlet coefficient subjected to the diffusion denoising processing in the third step by adopting Shearlet inverse transformation to obtain a reconstructed image;

fifthly, the image processor performs non-linear diffusion post-processing on the reconstructed image based on a partial differential equation to remove false details of the reconstructed image;

and step six, outputting the denoised image.

2. The Shearlet transform and nonlinear diffusion-based image denoising method according to claim 1, wherein: in the second step, the image processor carries out five-layer multi-scale Shearlet transformation on the noisy image, the two-level high-frequency sub-bands of the first layer and the second layer respectively have 16 directions, the two-level high-frequency sub-bands of the third layer and the fourth layer respectively have 8 directions, and the fifth layer is a low-frequency sub-band.

3. The Shearlet transform and nonlinear diffusion-based image denoising method according to claim 1, wherein: in the third step, the specific process of the image processor continuously using the lambda-order local wiener filtering to perform diffusion denoising processing on the initial noisy Shearlet coefficient is as follows: taking the initial noisy Shearlet coefficient as an input signal of first wiener filtering, and taking an output signal of current wiener filtering as an input signal of next wiener filtering; and recording the coefficient of the high-frequency Shearlet recovered by the r-th wiener filtering at the sub-band position (x, y) as z_r(x, y) and z_r(x,y)＝W_r(x,y)z_r-1(x, y), wherein r is a natural number from 1 to lambda, and z_r-1(x, y) is the high frequency Shearlet coefficient recovered at the sub-band position (x, y) with the r-1 th wiener filtering, z₀(x, y) is the initial noise-containing coefficient at the sub-band position (x, y), W_r(x, y) is the r-th wiener filter andn_r(x, y) is a noise component to be removed at the sub-band position (x, y),is n_rVariance of (x, y) and takenη is a noise variance adjustment parameter and η is a real number,is the initial noise variance within a subband andestimated by a Monte-Carlo method; s_r(x, y) is the signal component to be recovered at the sub-band position (x, y),is s is_rVariance of (x, y) andaccording to the formula <math> <mrow> <mi>E</mi> <mo>{</mo> <msubsup> <mi>s</mi> <mi>r</mi> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>}</mo> <mo>=</mo> <mi>E</mi> <mo>{</mo> <msubsup> <mi>z</mi> <mrow> <mi>r</mi> <mo>-</mo> <mn>1</mn> </mrow> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>}</mo> <mo>-</mo> <mi>&delta;</mi> <mi>E</mi> <mo>{</mo> <msubsup> <mi>n</mi> <mi>r</mi> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>,</mo> <mi>y</mi> <mo>)</mo> </mrow> <mo>}</mo> <mo>=</mo> <mfrac> <mn>1</mn> <msup> <mrow> <mo>(</mo> <mn>2</mn> <mi>R</mi> <mo>+</mo> <mn>1</mn> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mfrac> <munderover> <mo>&Sigma;</mo> <mrow> <mi>l</mi> <mn>1</mn> <mo>=</mo> <mo>-</mo> <mi>R</mi> </mrow> <mi>R</mi> </munderover> <munderover> <mo>&Sigma;</mo> <mrow> <mi>l</mi> <mn>2</mn> <mo>=</mo> <mo>-</mo> <mi>R</mi> </mrow> <mi>R</mi> </munderover> <msubsup> <mi>z</mi> <mrow> <mi>r</mi> <mo>-</mo> <mn>1</mn> </mrow> <mn>2</mn> </msubsup> <mrow> <mo>(</mo> <mi>x</mi> <mo>-</mo> <mi>l</mi> <mn>1</mn> <mo>,</mo> <mi>y</mi> <mo>-</mo> <mi>l</mi> <mn>2</mn> <mo>)</mo> </mrow> <mo>-</mo> <mi>&delta;</mi> <mi>&eta;</mi> <msup> <mover> <mi>&sigma;</mi> <mo>^</mo> </mover> <mn>2</mn> </msup> <mo>/</mo> <mi>&lambda;</mi> </mrow> </math> The estimation is carried out to obtain the result,is z_r-1Variance of (x, y) is s_r(x, y) is a non-negative real number, R is a constant and positive integer, l1 and l2 are integer variables and take values of-R to R, z_r-1(x-l1, y-l2) is the high frequency Shearlet coefficient recovered at the subband position (x-l1, y-l2) with the r-1 th wiener filtering, z-l 2₀(x-l1, y-l2) is the initial noise-containing coefficient at the subband position (x-l1, y-l 2).

4. The Shearlet transform and nonlinear diffusion-based image denoising method according to claim 3, wherein: the value of eta is a real number of 1-1.5.

5. The Shearlet transform and nonlinear diffusion-based image denoising method according to claim 1, wherein: in the step five, the image processor performs non-linear diffusion post-processing based on partial differential equation on the reconstructed image, and the specific process of removing the false details of the reconstructed image is as follows: assuming that the reconstructed image is I, partial differential equation is applied to ICarrying out nonlinear diffusion post-processing, wherein div is a divergence operator, t is the time of partial differential diffusion evolution, and the value of t is increased along with the depth of diffusion; i is_t(i, j) is the pixel grey value at location (i, j) at time t,is I_tGradient of (i, j), c_t(i, j) is an exponential diffusion function andpartial differential equation <math> <mrow> <mfrac> <mrow> <mo>&part;</mo> <msub> <mi>I</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> </mrow> <mrow> <mo>&part;</mo> <mi>t</mi> </mrow> </mfrac> <mo>=</mo> <mi>d</mi> <mi>i</mi> <mi>v</mi> <mo>&lsqb;</mo> <msub> <mi>c</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>&CenterDot;</mo> <mo>&dtri;</mo> <msub> <mi>I</mi> <mi>t</mi> </msub> <mrow> <mo>(</mo> <mi>i</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>&rsqb;</mo> </mrow> </math> Discretization is as follows: <math> <mrow> <msub> <mi>I</mi> <mi>k</mi> </msub> <mrow> <mo>(</mo> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> <mo>)</mo> </mrow> <mo>=</mo> <msub> <mi>I</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> <mo>)</mo> </mrow> <mo>+</mo> <mn>0.2</mn> <mo>&CenterDot;</mo> <munderover> <mo>&Sigma;</mo> <mrow> <mi>u</mi> <mo>=</mo> <mn>1</mn> </mrow> <mn>4</mn> </munderover> <mrow> <mo>(</mo> <mrow> <msubsup> <mi>I</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> <mi>u</mi> </msubsup> <mrow> <mo>(</mo> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mi>exp</mi> <mrow> <mo>&lsqb;</mo> <mrow> <mo>-</mo> <msup> <mrow> <mo>(</mo> <mrow> <mi>a</mi> <mi>b</mi> <mi>s</mi> <mrow> <mo>(</mo> <mrow> <msubsup> <mi>I</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> <mi>u</mi> </msubsup> <mrow> <mo>(</mo> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> <mo>)</mo> </mrow> <mo>-</mo> <msub> <mi>I</mi> <mrow> <mi>k</mi> <mo>-</mo> <mn>1</mn> </mrow> </msub> <mrow> <mo>(</mo> <mrow> <mi>i</mi> <mo>,</mo> <mi>j</mi> </mrow> <mo>)</mo> </mrow> </mrow> <mo>)</mo> </mrow> <mo>/</mo> <mi>K</mi> </mrow> <mo>)</mo> </mrow> <mn>2</mn> </msup> </mrow> <mo>&rsqb;</mo> </mrow> <mo>,</mo> </mrow> </math> wherein,(u-1, 2,3,4) is the gray value of the pixel positioned in the north, south, east and west four directions of (I, j), K is the gradient threshold value, the value of K is 1-10, and I_k(I, j) is the pixel gray value after k-th diffusion denoising at the position of (I, j), I_k-1(I, j) is the pixel gray value after k-1 diffusion denoising at the position of (I, j), I₀(I, j) is the pixel gray value of the reconstructed image I at the position (I, j), k is the number of iterations and k takes a positive integer between 1 and 7.