CN102890820B - Based on shearlet conversion and the image de-noising method of Wiener filtering - Google Patents

Abstract

The invention discloses a kind of based on shearlet conversion and the image de-noising method of Wiener filtering, the step of realization is: (1) input source image; (2) symmetric extension; (3) shear transformation; (4) WAVELET PACKET DECOMPOSITION; (5) Wiener filtering; (6) inverse wavelet package transforms; (7) inverse shear transformation; (8) inverse symmetry transformation; (9) image co-registration; (10) image after output denoising.Instant invention overcomes wavelet transformation in prior art and can not express the shortcoming of the anisotropy information of image very well, and the problem that the denoising effect using single threshold value to carry out same treatment to coefficient on different directions and cause is undesirable, make use of shearlet conversion there is multidirectional and Wiener filtering according to advantages such as the Local Deviation adjustment wave filter outputs of image, thus can to analyze image detail information more accurately in the high frequency coefficient on the different directions of image.Finally obtain image after high-quality denoising.

Description

Based on shearlet conversion and the image de-noising method of Wiener filtering

Technical field

The invention belongs to technical field of image processing, further relate to the image de-noising method based on shearlet conversion and Wiener filtering in Image semantic classification field.Can be applicable to the optics gray level image denoising containing white Gaussian noise, to obtain the image more clearly with high s/n ratio.The present invention is applied in graphical analysis, Image semantic classification the noise that can effectively reduce in image, especially for the image containing Gaussian noise, can obtain better denoising effect, more can meet the visual psychology of people and the requirement of practical application.

Background technology

In Image semantic classification field, in order to remove the white Gaussian noise contained in original image, to obtain the picture rich in detail of high-quality, high s/n ratio, and for post processing of image provides advantage and adopt the method for image denoising.Current image de-noising method mainly adopts the method such as the Threshold Denoising Method based on wavelet transformation and the Wiener filtering denoising based on wavelet transformation to realize image denoising.

Patented technology " a kind of method for de-noising dual-tree complex wavelet image based on the partial differential equation " (publication number: CN101777179A that University of Electronic Science and Technology has, authorize day: on 02 15th, 2012, applying date: on 02 05th, 2010) in disclose a kind of method for de-noising dual-tree complex wavelet image based on partial differential equation.First the method carries out dual-tree complex wavelet transform decomposition to the noisy image of input, and carries out isotropic diffusion to two low frequency sub-band images after decomposing; The adaptive model that bamboo product improves, calculate dual-tree complex wavelet transform mould and the gradient-norm of high frequency detail sub-band images on each direction, utilize the weighted mean of two tree WAVELET TRANSFORM MODULUS and gradient-norm to design a kind of adaptive coefficient of diffusion function to improve P-M model; Then to the adaptive model sliding-model control improved, and anisotropy parameter is carried out to six high-frequency sub-band images; Finally carry out dual-tree complex wavelet inverse transformation, export the image after denoising.Although the method has the ability distinguishing noise and signal preferably, the shortcoming still existed is, dual-tree complex wavelet transform lacks translation invariance, and after causing the denoising obtained, distortion appears in image, and main manifestations is ringing effect and pseudo-Gibbs effect.In addition the method does not consider the difference of noise on image annoyance level on the different scale of wavelet decomposition when denoising, therefore can not reach good denoising effect.

The people such as Tian Pei propose the image de-noising method based on wavelet transformation and Wiener filtering in document " Tian Pei; Li Qingzhou; Ma Ping; Niu Yuguang; ' a kind of image denoising new method based on wavelet transformation ' [J]; Chinese graphics image journal, 13 (3), 395-399 (2008) ".First the method carries out wavelet transformation to image; Then according to the different characteristics existed between the wavelet coefficient of Gaussian noise and the wavelet coefficient of image, Wiener filtering is carried out to the wavelet coefficient on different scale different directions; Finally inverse wavelet transform is carried out to filtered wavelet coefficient, obtain image after denoising.Although the method can improve the peak value to-noise ratio (PSNR) of image, and can retain more image detail information.But the shortcoming still existed is, wavelet transformation well can not express the anisotropic detailed information in image, therefore can not remove the noise contained in anisotropic image well.

In sum, although the image de-noising method based on wavelet transformation obtains good effect in image denoising.But wavelet transformation well can not express the anisotropy information of image, as the marginal information in digital picture and line feature information, so after utilizing the denoising obtained based on the denoising method of wavelet transformation in image, the inevitable blooming occurred to a certain degree on the edge in image and details position.

Summary of the invention

The shortcoming of the undesirable and image fault of the image denoising effect that the object of the invention is to cause based on the image de-noising method of wavelet transformation for prior art, proposes the image de-noising method based on shearlet conversion and Wiener filtering.In the present invention symmetric extension is included, three parts of shear transformation and WAVELET PACKET DECOMPOSITION to the decomposition principle based on shearlet conversion of image.The present invention makes full use of shearlet transfer pair image and carries out decomposing the advantage that the coefficient of dissociation obtained better can show the detailed information of image, utilize Wiener filtering better can remove the feature of white Gaussian noise simultaneously, shearlet conversion and Wiener filtering are combined to carry out image denoising.After the denoising finally obtained, image can restraint speckle effectively, can retain again the more detailed information of image.

The concrete steps that the present invention realizes are as follows:

(1) input source image

Apply matlab software in a computer and read the source images be stored in hard disc of computer space.

(2) symmetric extension

2a) horizontal symmetrical expansion is carried out to source images, with a certain bar in two of image vertical edges boundary lines for axis of symmetry, by horizontal extension formula by the another side of image mapped to axis of symmetry, obtain horizontal extension image;

2b) vertical symmetry expansion is carried out to source images, with a certain bar in two of image horizontal sides boundary lines for axis of symmetry, by extends perpendicular formula by the another side of image mapped to axis of symmetry, obtain extends perpendicular image.

(3) shear transformation

By shear transformation formula, shear transformation is carried out to image after the expansion obtained in step (2), and by the image after shear transformation stored in calculator memory.

(4) WAVELET PACKET DECOMPOSITION

Utilize discrete wavelet packet disassembling tool to carry out wavelet packet multi-resolution decomposition to image after shear transformation respectively, obtain the low frequency coefficient after decomposing and high frequency coefficient, stored in calculator memory.

(5) Wiener filtering

5a) all in read step (4) high frequency coefficients;

5b) utilize S filter instrument to step 5a) in read high frequency coefficient carry out Wiener filtering process, obtain filtered high frequency coefficient, by filtered high frequency coefficient stored in calculator memory.

(6) inverse wavelet package transforms

Utilize inverse wavelet package transforms instrument to step 5b) in the low frequency coefficient of correspondence that obtains in the filtered high frequency coefficient that obtains and step (4) carry out inverse wavelet package transforms, obtain the image after inverse wavelet package transforms, and stored in calculator memory.

(7) inverse shear transformation

By inverse shear transformation formula, inverse shear transformation is carried out to the image after the inverse wavelet package transforms obtained in step (6), and by image after the inverse shear transformation that obtains stored in calculator memory.

(8) inverse symmetric extension

Level is carried out against symmetric extension and vertical inverse symmetric extension by level against symmetric extension formula and vertical inverse symmetric extension formula respectively to the image obtained in step (7), obtains the rear image of inverse expansion.

(9) image co-registration

By data average formula, image co-registration is carried out to image after the inverse expansion of step (8), obtains the image after denoising.

(10) result images after denoising is exported.

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

First, the present invention exploded view as time, application shearlet transfer pair image carries out multi-resolution decomposition, has multidirectional due to shearlet conversion, thus high-frequency information and the low-frequency information of Noise image can be obtained in a plurality of directions, effectively to catch image detail.Overcome wavelet transformation in prior art and can not express the shortcoming of the anisotropy information of image very well, the detailed information of the Noise image using the method in the present invention to obtain can be analyzed more accurately.

The second, the present invention, when carrying out filtering process to the high frequency coefficient after noisy picture breakdown, applies the method for Wiener filtering.Because Wiener filtering can adjust the output of wave filter according to the Local Deviation of image, to carry out region adaptivity filtering process to the high frequency coefficient of Noise image.Overcome the problem that Threshold Denoising Method of the prior art carries out identical process to the coefficient on different scale different directions and the denoising effect that causes is undesirable, white Gaussian noise in the high frequency coefficient of the Noise image using the method in the present invention to obtain can better be suppressed.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention;

Fig. 2 is the present invention for the analogous diagram of being carried out denoising by the noisy image of noise slight pollution;

Fig. 3 is the present invention for the analogous diagram of being carried out denoising by the noisy image of noise severe contamination.

Embodiment

With reference to Fig. 1, the specific embodiment of the invention is as follows:

Step 1, input source image

Apply matlab software in a computer and read the source images be stored in hard disc of computer space.

Step 2, symmetric extension

In order to avoid the boundary effect that the image after shear transformation may occur in denoising process, the present invention is to first carrying out symmetric extension in both the horizontal and vertical directions respectively to source images.

2a) horizontal symmetrical expansion is carried out to source images, with a certain bar in two of image vertical edges boundary lines for axis of symmetry, by horizontal extension formula by the another side of image mapped to axis of symmetry, obtain horizontal extension image;

Wherein, f (i, j) is for source images is at the gray-scale value of coordinate (i, j) position, and n is the columns of the pixel of source images, f _h(i, j) is for horizontal extension image is at the gray-scale value of coordinate (i, j) position;

2b) vertical symmetry expansion is carried out to source images, with a certain bar in two of image horizontal sides boundary lines for axis of symmetry, by extends perpendicular formula by the another side of image mapped to axis of symmetry, obtain extends perpendicular image;

Wherein, f (i, j) is for source images is at the gray-scale value of coordinate (i, j) position, and m is the line number of the pixel of source images, and f (i, j) is for horizontal extension image is at the gray-scale value of coordinate (i, j) position.

Step 3, shear transformation

By shear transformation formula, shear transformation is carried out to image after expansion, and by the image after shear transformation stored in calculator memory;

if

if

Wherein, x ', y ' be coordinate corresponding to image pixel after shear transformation, x, y are the coordinate that after symmetric extension, image pixel is corresponding, k ∈ [-2 ^(ndir), 2 ^(ndir)], k ∈ Z, ndir are direction parameter, ndir=0,1 or 2, s shears matrix, f ' (x ', y ') for image after shear transformation coordinate (x ', y ') grey scale pixel value of position, f (x, y) is for transform expansion image is at the gray-scale value of coordinate (x, y) position.

Step 4, WAVELET PACKET DECOMPOSITION

Utilize discrete wavelet packet disassembling tool to carry out three layers of wavelet packet multi-resolution decomposition to image after shear transformation respectively, obtain the low frequency coefficient after decomposing and high frequency coefficient, stored in calculator memory;

4a) by the low-pass filter in image input WAVELET PACKET DECOMPOSITION instrument after shear transformation, obtain one deck low frequency coefficient I of WAVELET PACKET DECOMPOSITION through filtering _l1;

4b) by the Hi-pass filter in image input WAVELET PACKET DECOMPOSITION instrument after shear transformation, obtain one deck high frequency coefficient I of WAVELET PACKET DECOMPOSITION through filtering _h1;

One deck low frequency coefficient I 4c) WAVELET PACKET DECOMPOSITION obtained _l1low-pass filter in input WAVELET PACKET DECOMPOSITION instrument, obtains two layers of low frequency coefficient I of WAVELET PACKET DECOMPOSITION through filtering _l2;

One deck low frequency coefficient I 4d) WAVELET PACKET DECOMPOSITION obtained _l1hi-pass filter in input WAVELET PACKET DECOMPOSITION instrument, obtains two layers of high frequency coefficient I of WAVELET PACKET DECOMPOSITION through filtering _h2;

Two layers of low frequency coefficient I 4e) WAVELET PACKET DECOMPOSITION obtained _l2low-pass filter in input WAVELET PACKET DECOMPOSITION instrument, obtains three layers of low frequency coefficient I of WAVELET PACKET DECOMPOSITION through filtering _l3;

Two layers of low frequency coefficient I 4f) WAVELET PACKET DECOMPOSITION obtained _l2hi-pass filter in input WAVELET PACKET DECOMPOSITION instrument, obtains three layers of high frequency coefficient I of WAVELET PACKET DECOMPOSITION through filtering _h3;

Step 5, Wiener filtering

5a) read WAVELET PACKET DECOMPOSITION and obtain all high frequency coefficients;

5b) utilize S filter instrument to carry out Wiener filtering process to the high frequency coefficient read, obtain filtered high frequency coefficient, the window size of S filter selects 5 × 5, then by filtered high frequency coefficient stored in calculator memory.

Step 6, inverse wavelet package transforms

Utilize inverse wavelet package transforms instrument to step 5b) in the low frequency coefficient of correspondence that obtains in the filtered high frequency coefficient that obtains and step 4 carry out inverse wavelet package transforms, obtain the image f after inverse wavelet package transforms " (x ", y "), and stored in calculator memory.

Step 7, inverse shear transformation

By inverse shear transformation formula, inverse shear transformation is carried out to the image after the inverse wavelet package transforms obtained in step (6), and by image after the inverse shear transformation that obtains stored in calculator memory;

if

if

Wherein, x _i", y _i" be the coordinate that after inverse shear transformation, image pixel is corresponding, subscript i represents the inverse transformation of shear transformation, x ", y " be the coordinate that image pixel is corresponding, k ∈ [-2 ^(ndir), 2 ^(ndir)], k ∈ Z, superscript ndir are direction parameter, ndir=0,1 or 2, s ' inverse shear matrix, f " (x ", y ") be before shear transformation image (x ", the gray-scale value of y ") position, f _i" (x _i", y _i") be after inverse shear transformation image at (x _i", y _ithe gray-scale value of ") position.

Step 8, inverse symmetric extension

Level is carried out against symmetric extension and vertical inverse symmetric extension by level against symmetric extension formula and vertical inverse symmetric extension formula respectively to the image obtained in step (7), obtains the rear image of inverse expansion;

Level is as follows against symmetric extension formula:

$f(i,j)=\frac{f(i,j)+f(i,2n-j)}{2},(1\≤i\≤m,1\≤j\≤n)$

Wherein, f (i, j) is for level is against the gray-scale value of the image pixel after symmetric extension on coordinate (i, j) position, m and n is respectively line number and the columns of source images.

Vertical inverse symmetric extension formula is as follows:

$f(i,j)=\frac{f(i,j)+f(2m-i,j)}{2},(1\≤i\≤m,1\≤j\≤n)$

Wherein, f (i, j) is the gray-scale value of image pixel on coordinate (i, j) position after vertical inverse symmetric extension, m and n is respectively line number and the columns of source images.

Step 9, image co-registration

By data average formula, image co-registration is carried out to image after the inverse expansion of step (8), obtains the image after denoising;

$F(i,j)=\frac{\underset{p=1}{\overset{\mathrm{num}}{\mathrm{\Σ}}}{f}_p(i,j)}{\mathrm{num}}$

Wherein, F (i, j) is for image after denoising is at the gray-scale value of coordinate (i, j) position, and num is the number of the image after inverse expansion, f _p(i, j) is p the inverse gray-scale value of expanded images in coordinate (i, j) position.

Step 10, exports the result images after denoising.

Effect of the present invention can be further illustrated by following emulation.

Emulation 1, in the present invention by the emulation of the noisy image denoising of noise slight pollution.

Emulating 1 simulated conditions is carry out under MATLAB7.0 software.

With reference to Fig. 2, to conventional Lena image, size is 512 × 512 pixels, and 256 grades of gray level images carry out emulation experiment.Fig. 2 (a) is the original image of Lena clearly; Fig. 2 (b) be containing noise criteria difference be 50 Lena image; Fig. 2 (c) is the denoising result based on the inventive method; Fig. 2 (d) is the result of the Wiener filtering denoising based on contourlet transformation; Fig. 2 (e) is the result of the Wiener filtering denoising based on wavelet transformation.

As can be seen from Fig. 2 (e), the Partial Feature of image based on missing image after the denoising that the Wiener filtering denoising method (WTWN) of wavelet transformation obtains, and also after denoising, image is smooth-out, causes image fault.Can clearly find out from Fig. 2 (d), based on the Wiener filtering denoising method (CTWN) of contourlet transformation although image can keep the marginal information of image preferably after the denoising that obtains, the subjective quality assessment of image declines to some extent.And clearly can find out that the denoising method (STWN) that the present invention proposes can obtain good denoising effect from Fig. 2 (c), not only restrained effectively noise, improve signal noise ratio (snr) of image; And well remain the detailed information of image.Respectively following evaluating is calculated to the denoising image that three kinds of methods obtain above: Y-PSNR (PSNR), final data is as shown in table 1.

Show the PSNR value after 1CTWN, STWN and the inventive method denoising

σ	noisy	WTWN	CTWN	STWN
					10	28.1179	34.7788	33.9715	35.167
20	22.1020	30.8331	30.6299	31.8370
					30	18.5790	28.3909	28.9229	30.108
40	16.0851	26.5236	27.8225	28.964
					50	14.1275	25.0222	26.8859	28.004
60	12.5705	23.6963	26.2903	27.356

As can be seen from the objective evaluation measured value of table 1, the present invention is better than other all algorithms existing, and Y-PSNR is the highest, and show that the present invention can obtain better denoising effect, after denoising, image is more clear, has higher PSRN.

Emulation 2, in the present invention by the emulation of the noisy image denoising of noise severe contamination.

Emulating 2 simulated conditions is carry out under MATLAB7.0 software.

With reference to Fig. 3, to conventional Lena image, size is 512 × 512 pixels, and 256 grades of gray level images carry out emulation experiment.Fig. 3 (a) is the original image of Lena clearly; Fig. 3 (b) be containing noise criteria difference be 100 Lena image; Fig. 3 (c) is the denoising result based on the inventive method; Fig. 3 (d) is the result of hard-threshold (STTD) denoising based on shearlet conversion.

From passable the finding out of Fig. 3 (d), based on shearlet conversion hard-threshold denoising method (STTD) although denoising after image can remain the local features such as image border preferably, there is ring, pseudo-Gibbs vision distortion phenomenon.And can find out that denoising method (STWN) that the present invention proposes not only can restraint speckle effectively from Fig. 3 (c), and better stay the detailed information of image.Respectively following evaluating is calculated to the denoising image that two kinds of methods obtain above: Y-PSNR (PSNR), final data is as shown in table 2.

Show the PSNR value after 2STTD and the inventive method denoising

σ	noisy	STTD	STWN
				40	16.1080	28.7026	28.9138
50	14.1409	27.5368	28.0370
				60	12.5616	26.6703	27.3638
70	11.2193	25.8510	26.7635
				80	10.0596	25.2345	26.3891
90	9.0459	24.6395	25.9372
				100	8.1363	24.1178	25.5594
150	4.6131	21.9662	24.2220
				200	2.1085	20.2769	23.2871
250	0.1905	19.1194	22.6823

As can be seen from the objective evaluation measured value of table 2, the present invention is better than the hard-threshold denoising method based on shearlet conversion, after denoising, the PSNR of image and edge hold facility have more advantage than the hard-threshold denoising based on shearlet conversion, and can find out that this advantage is more and more obvious along with image contains the increase of noise criteria difference.The denoising effect figure of two kinds of distinct methods in the PSNR value obtained by two kinds of distinct methods in respectively comparison sheet 2 and Fig. 3, we can show that the present invention can obtain better denoising effect, obtain higher PSRN, the information of better reservation image.

Claims (10)

1., based on shearlet conversion and the image de-noising method of Wiener filtering, concrete operation step is as follows:

(1) input source image

Apply matlab software in a computer and read the source images be stored in hard disc of computer space;

(2) symmetric extension

2a) horizontal symmetrical expansion is carried out to source images, with a certain bar in two of image vertical edges boundary lines for axis of symmetry, by horizontal extension formula by the another side of image mapped to axis of symmetry, obtain horizontal extension image;

2b) vertical symmetry expansion is carried out to source images, with a certain bar in two of image horizontal sides boundary lines for axis of symmetry, by extends perpendicular formula by the another side of image mapped to axis of symmetry, obtain extends perpendicular image;

(3) shear transformation

By shear transformation formula, shear transformation is carried out to image after the expansion obtained in step (2), and by the image after shear transformation stored in calculator memory;

(4) WAVELET PACKET DECOMPOSITION

Utilize discrete wavelet packet disassembling tool to carry out wavelet packet multi-resolution decomposition to image after shear transformation respectively, obtain the low frequency coefficient after decomposing and high frequency coefficient, stored in calculator memory;

(5) Wiener filtering

5a) all in read step (4) high frequency coefficients;

5b) utilize S filter instrument to step 5a) in read high frequency coefficient carry out Wiener filtering process, obtain filtered high frequency coefficient, by filtered high frequency coefficient stored in calculator memory;

(6) inverse wavelet package transforms

Utilize inverse wavelet package transforms instrument to step 5b) in the low frequency coefficient of correspondence that obtains in the filtered high frequency coefficient that obtains and step (4) carry out inverse wavelet package transforms, obtain the image after inverse wavelet package transforms, and stored in calculator memory;

(7) inverse shear transformation

By inverse shear transformation formula, inverse shear transformation is carried out to the image after the inverse wavelet package transforms obtained in step (6), and by image after the inverse shear transformation that obtains stored in calculator memory;

(8) inverse symmetric extension

Level is carried out against symmetric extension and vertical inverse symmetric extension by level against symmetric extension formula and vertical inverse symmetric extension formula respectively to the image obtained in step (7), obtains the rear image of inverse expansion;

(9) image co-registration

By data average formula, image co-registration is carried out to image after the inverse expansion of step (8), obtains the image after denoising;

(10) result images after denoising is exported.

2. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, to it is characterized in that: step 2a) described in horizontal symmetrical extends equation as follows:

Wherein, f (i, j) is for source images is at the gray-scale value of coordinate (i, j) position, and n is the columns of the pixel of source images, f _h(i, j) is for horizontal extension image is at the gray-scale value of coordinate (i, j) position.

3. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, to it is characterized in that: step 2b) described in vertical symmetry extends equation as follows:

Wherein, f (i, j) is for source images is at the gray-scale value of coordinate (i, j) position, and m is the line number of the pixel of source images, f _v(i, j) is for horizontal extension image is at the gray-scale value of coordinate (i, j) position.

4. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, it is characterized in that: the shear transformation formula described in step (3) is as follows:

if

if

Wherein, (x ', y ') is coordinate corresponding to image pixel after shear transformation, and (x, y) is coordinate corresponding to image pixel after symmetric extension, k ∈ [-2 ^(ndir), 2 ^(ndir)], k ∈ Z, represent and round downwards, ndir is direction parameter, ndir=0,1 or 2, s be shear matrix, f ' (x ', y ') for image after shear transformation is at the grey scale pixel value of coordinate (x ', y ') position, f (x, y) for transform expansion image is at the gray-scale value of coordinate (x, y) position.

5. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, it is characterized in that: the WAVELET PACKET DECOMPOSITION described in step (4) adopts three layers of WAVELET PACKET DECOMPOSITION.

6. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, to it is characterized in that: step 5b) described in high frequency coefficient Wiener filtering in the window size selection 5 × 5 of S filter.

7. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, it is characterized in that: the inverse shear transformation formula described in step (7) is as follows:

if

if

Wherein, (x _i", y _i") is the coordinate that after inverse shear transformation, image pixel is corresponding, and subscript i represents the inverse transformation of shear transformation, (x ", y ") is coordinate corresponding to image pixel, k ∈ [-2 ^(ndir), 2 ^(ndir)], k ∈ Z, represent and round downwards, superscript ndir is direction parameter, ndir=0,1 or 2, s ' inverse shear matrix, f " (x ", y ") be before shear transformation image (x ", the gray-scale value of y ") position, f _i" (x _i", y _i") be after inverse shear transformation image at (x _i", y _ithe gray-scale value of ") position.

8. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, it is characterized in that: the level described in step (8) is as follows against symmetric extension formula:

$f(i,j)=\frac{f(i,j)+f(i,2n-j)}{2},1\≤i\≤m,1\≤j\≤n$

Wherein, f (i, j) is for level is against the gray-scale value of the image pixel after symmetric extension on coordinate (i, j) position, m and n is respectively line number and the columns of source images.

9. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, it is characterized in that: vertically as follows against symmetric extension formula described in step (8):

$f(i,j)=\frac{f(i,j)+f(2m-i,j)}{2},1\≤i\≤m,1\≤j\≤n$

Wherein, f (i, j) is the gray-scale value of image pixel on coordinate (i, j) position after vertical inverse symmetric extension, m and n is respectively line number and the columns of source images.

10. according to claim 1 based on shearlet conversion and the image de-noising method of Wiener filtering, it is characterized in that: the data average formula described in step (9) is as follows:

$F(i,j)=\frac{\underset{p=1}{\overset{\mathrm{num}}{\mathrm{\Σ}}}{f}_p(i,j)}{\mathrm{num}}$

Wherein, F (i, j) is for image after denoising is at the gray-scale value of coordinate (i, j) position, and num is the number of the image after inverse expansion, f _p(i, j) is p the inverse gray-scale value of expanded images in coordinate (i, j) position.