CN105427249A - Wind power image quality enhancing method based on robustness nuclear norm regular regression - Google Patents

Abstract

The invention discloses a wind power image quality enhancing method based on robustness nuclear norm regular regression. The method comprises steps that firstly, each pixel position of an image is taken as a center, and image blocks of all pixel positions of a low quality test image and a training sample image are acquired; secondly, for all image blocks of the lower quality test image, the robustness nuclear norm regular regression method is utilized to acquire robustness linearity expression of an image block set of all image blocks on the corresponding positions of the lower quality training sample image; thirdly, on the condition that the robustness expression coefficient is not changed, high quality training sample image blocks are utilized to replace low quality training sample image blocks, so the high quality test image blocks corresponding to low quality test image blocks are acquired, non-local self-similarity is utilized to realize further quality enhancement for the high quality test image blocks, and fourthly, the high quality test image blocks after further quality enhancement are connected in series and are integrated, and the high quality test image is acquired.

Description

A kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical

Technical field

The present invention relates to a kind of method for enhancing image quality, particularly a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical, belongs to technical field of image processing.

Background technology

Image recognition is generally divided into four steps: Image semantic classification, feature extraction, feature representation and classification.In order to follow the tracks of and identify the running status of pneumatic equipment blades better, usually need to take some Wind turbines images.The image of traditional images recognizer supposition input has good resolution.But in real life, because target image is usually distant with picture pick-up device, be subject to again the change of illumination condition, the impact of the factors such as the noise of the motion blur of target image and equipment self, the image obtained often size is little, resolution is low, and noise is large, and characteristic details information is also extremely limited.In this case, the recognition effect that traditional image recognition algorithm obtains does not reach the result that people expect far away.

By method for enhancing image quality, the performance of recognizer effectively can be improved.It is one of study hotspot in Digital Image Processing that single image quality strengthens technology, and a large amount of researchist has both at home and abroad carried out extensive and deep research to it.From the angle of historical development, picture quality strengthens technology and roughly experienced by interpolation method, method for reconstructing and machine learning method three phases.Conventional super-resolution treatment technology based on interpolation generally regards discrete picture signal as continuous signal, utilizes continuity Characteristics to estimate the pixel value of unknown sampled point.But the natural image in real world contains a large amount of singularity characteristics, such as edge, angle point and high frequency texture area etc.Therefore, the interpolation algorithm based on continuity hypothesis often forms obvious ring and aliasing effect, has a strong impact on superresolution processing effect.Based on the forming process of the method simulate digital images of rebuilding, solve net result by process model building digital imaging apparatus being gathered to picture signal.This method often needs to suppose some parameter, point spread function (the PointSpreadFunction of such as video camera, PSF), environmental noise etc., but the parameter of these hypothesis is not inconsistent with actual parameter under normal circumstances, even differ greatly, this can cause super-resolution result to occur severe deviations.In addition, very responsive to amplification factor is also one of major defect of method for reconstructing, and this can have a strong impact on the application of algorithm.Comparatively speaking, machine learning method is more attractive than the above two, is no matter angle from theoretical research or the angle from practice principle is obtained for good development.

Current method for enhancing image quality is sorted out as follows:

1. neighborhood embedding inlay technique (NE)

H.Chang, D.Y.Yeung and Y.M.Xiong in 2004 at IEEEConferenceonComputerVisionandPatternRecognition (pp.1275-1282, vol.1) in " Super-resolutionthroughneighborembedding " that deliver, based on local linear embedding theory, by hypothesis low resolution feature space and high-resolution features space, there is analog structure and reduce reconstruction sample quantity, while maintenance Similarity Structure, use high-resolution features to replace low resolution feature, obtain high-quality image block.While raising algorithm execution efficiency, obtain good picture quality and strengthen the property.

2. eigentransformation method (Eigentransformation)

X.Wang, X.Tang in 2005 at IEEETransactionsonSystems, Man, andCybernetics (PartC:ApplicationandReviews, vol.35, no.3, pp.425-434), in " Hallucinatingfacebyeigentransformation " that deliver, principal component analysis (PCA) (PCA) is used to carry out the structural similarity of picture engraving.PCA is used for low resolution input picture to be expressed as the linear combination that low resolution trains dictionary.By low resolution training dictionary is replaced to corresponding high resolving power training dictionary, high-resolution image can be obtained.

3. sparse representation method (SR)

J.Yang, J.Wright, H.Tang and Y.Ma in 2010 at IEEETransactionsonImageProcessing (vol.19, no.11, pp.2861-2873), in " Imagesuper-resolutionviasparserepresentation " that deliver, proposition calculates test pattern block with sparse expression and represent coefficient in training image blocks.Owing to not fixing neighborhood quantity in sparse approximation, but make approximate error little as far as possible, effectively prevent the inappropriate fitting problems in coefficient solution procedure.

Above method, before solving expression coefficient, is all wanted in advance two-dimentional low-resolution image (block) to be expressed as a column vector.This inevitably loses some structural informations (such as rank of matrix) of two dimensional image, makes to represent that solving of coefficient is not very accurate.And in previous methods, the model construction under noise situations is not discussed.

Summary of the invention

Technical matters to be solved by this invention is to provide a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical, for existing method to this shortcoming of the noise-sensitive that may contain in test pattern, meet the demand to high quality graphic in practical application.

The present invention is for solving the problems of the technologies described above by the following technical solutions:

The invention provides a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical, comprise following concrete steps:

Step 1, in image centered by each location of pixels, obtains the image block of inferior quality test pattern and each location of pixels of training sample image;

Step 2, to each image block in inferior quality test pattern, uses robust nuclear norm canonical homing method to obtain the robust linear expression of its image block set in low quality training sample image on correspondence position;

Step 3, when keeping robust representation coefficient constant, low quality training sample image block is replaced with high quality training sample image block, thus obtain high-quality test pattern block corresponding to inferior quality test pattern block, more non local self-similarity is used to carry out further quality enhancing to high-quality test pattern block;

Step 4, connects to the high-quality test pattern block that quality further in step 3 strengthens and integrates, thus obtaining high-quality test pattern.

As further prioritization scheme of the present invention, centered by each location of pixels, obtain the image block of inferior quality test pattern and each location of pixels of training sample image, be specially in image in step 1:

1.1) lower quality version of training image is obtained;

1.2) in image, centered by each location of pixels, the image block of inferior quality test pattern and each location of pixels of training sample image is obtained.

As further prioritization scheme of the present invention, by the lower quality version of twice bicubic interpolation acquisition training image in step 1.1.

As further prioritization scheme of the present invention, to each image block in inferior quality test pattern in step 2, use robust nuclear norm canonical homing method to obtain the robust linear expression of its image block set in low quality training sample image on correspondence position, be specially:

D＝x ₁C ₁+x ₂C ₂+…+x _MC _M+e

Wherein, D is the inferior quality test pattern block of location of pixels (i, j); C _rthe image block of position (i, j) in r inferior quality pixel training sample, r={1,2 ..., M}, M are training sample numbers; x _rfor representing r element value in coefficient vector x;

Based on robust regression thought, described robust linear expression coefficient vector x obtains according to following model solution:

$\underset{x}{\min }\left|\right|x|{|}_q+\mathrm{\λ}|\left|Wo\right(C\left(x\right)-D\left)\right|{|}_{*}$

In formula, || x|| _qthe q norm representing coefficient vector x, λ is the first regularization parameter; W is weight matrix, the i-th row jth column element of W α and β is the parameter pre-set, Y _i,jfor representing the i-th row jth column element value of residual matrix Y; The Hadamard of ο representing matrix amasss; C (x)=x ₁c ₁+ x ₂c ₂+ ... + x _mc _mfrom space arrive a linear mapping, d × d is the size of image block; || || _*represent nuclear norm.

As further prioritization scheme of the present invention, according to model the method solving robust linear expression coefficient vector x is as follows:

2.1) set greatest iteration coefficient, put primary iteration number of times t=0;

2.2) upgrade weight matrix W and represent residual matrix Y, being specially:

$W_{i,j}=\frac{\exp \left(\mathrm{\α}\mathrm{\β}-\mathrm{\α}{\left(Y_{i,j}\right)}^2\right)}{1+xp\left(\mathrm{\α}\mathrm{\β}-\mathrm{\α}{\left(Y_{i,j}\right)}^2\right)}$

Y＝D-D ^(t)

Wherein, D ^(t)the inferior quality test pattern block of location of pixels (i, j) when being the t time iteration, and D ⁽⁰⁾=D _ini, D _inithe average image of low quality training sample,

2.3) upgrade inferior quality test pattern block and low quality training image block, be specially:

A＝WoD

B _r＝WοC _r

Wherein, A is the inferior quality test pattern block of location of pixels (i, j) after weighting, B _rit is the image block of location of pixels (i, j) in r low quality training sample after weighting; 2.4) Renewal model be specially:

$\underset{x,z,E}{\min }\left|\right|z|{|}_q+\mathrm{\λ}|\left|E\right|{|}_{*}$

s.t.G(x)-A＝Eandx＝z

Its Lagrangian function is expressed as:

$L_{\mathrm{\μ}}(x,z,E)=\left|\right|z|{|}_q+\mathrm{\λ}|\left|E\right|{|}_{*}+\frac{\mathrm{\μ}}{2}\left(\right||x-z+\frac{1}{\mathrm{\μ}}{y}_1|{|}_F^2+\left|\right|G\left(x\right)-A-E+\frac{1}{\mathrm{\μ}}{Y}_2\left|{|}_F^2\right)-\frac{1}{2\mathrm{\μ}}\left|\right|y_1|{|}_2^2-\frac{1}{2\mathrm{\μ}}|\left|Y_2\right|{|}_F^2$

Wherein, y ₁, Y ₂be Lagrange multiplier, μ is the second regularization parameter, for the F norm of matrix;

2.5) alternating direction multiplier method ADMM is adopted to solve the model after renewal in step 2.5, the expression coefficient x exported when obtaining the t time iteration _r ^(t)thus the expression coefficient vector x exported when obtaining the t time iteration ^(t);

2.6) reconstruct inferior quality test pattern block, be specially:

$D^{\left(t\right)}={\Σ}_{r=1}^M{x_r}^{\left(t\right)}{B}_r$

Wherein, D ^(t)inferior quality test pattern block when being the t time iteration;

2.7) if reach maximum iteration time or following end condition, export and represent coefficient vector x ^(t)as final solving result; Otherwise, put t=t+1, return step 2.2;

||W ^(t)-W ^(t-1)|| ₂/||W ^(t-1)|| ₂≤γ

Wherein, γ is the first default fault-tolerant value.

As further prioritization scheme of the present invention, in step 2.2, maximum iteration time is more than or equal to 2.

As further prioritization scheme of the present invention, adopt alternating direction multiplier method ADMM to solve the model in step 2.5 in step 2.5, be specially:

2.5.1) fixing x, E, upgrades z, is specially:

A) as q=1, z is tried to achieve by soft threshold method:

$z^{l+1}=\underset{z}{\arg \min }\frac{1}{\mathrm{\μ}}\left|\right|z|{|}_1+\frac{1}{2}||z-\left(x^l+\frac{1}{\mathrm{\μ}}{y}_1^l\right)|{|}_F^2;$

B) as q=2,

$z^{l+1}=\frac{\mathrm{\μ}}{\mathrm{\μ}+2}\left(x^l+\frac{1}{\mathrm{\μ}}{y}_1^l\right);$

Wherein, z ^l+1be the value of z after l+1 step upgrades, x ^l, y ₁ ^lbe respectively l step and upgrade rear x, y ₁value;

2.5.2) fixing z, E, upgrades x, is specially:

$x^{l+1}=(H^{T}H+I)\backslash \[H^{T}Vec\left(A+E^l-\frac{1}{\mathrm{\μ}}{Y}_2^l\right)+z^{l+1}-\frac{1}{\mathrm{\μ}}{y}_1^l\];$

Wherein, x ^l+1be the value of x after l+1 step upgrades, H=[Vec (B ₁) ..., Vec (B _m)], the vectorization operation of Vec () representing matrix, E ^l, y ₁ ^l, Y ₂ ^lbe respectively l step and upgrade rear E, y ₁, Y ₂value;

2.5.3) fixing x, z, upgrades E, is specially:

Optimum E is solved by singular value thresholding ^l+1:

$E^{l+1}=\underset{E}{\arg \min }\frac{\mathrm{\λ}}{\mathrm{\μ}}\left|\right|E|{|}_{*}+\frac{1}{2}||E-(G\left(x^{l+1}\right)-A+\frac{1}{\mathrm{\μ}}{Y}_2^l\left)\right|{|}_F^2;$

Wherein, E ^l+1be the value of E after l+1 step upgrades,

2.5.4) upgrade Lagrange multiplier, be specially:

$y_1^{l+1}=y_1^l+\mathrm{\μ}(x^{l+1}-z^{l+1})$

$Y_2^{l+1}=Y_2^l+\mathrm{\μ}(G\left(x^{l+1}\right)-A-E^{l+1})$

Wherein, be respectively l+1 step and upgrade rear y ₁, Y ₂value;

2.5.5) if reach following end condition, then x is exported ^l+1as x _r ^(t); Otherwise, turn back to step 2.5.1;

$\left|\right|G\left(x\right)-A-E+\frac{1}{\mathrm{\μ}}{Y}_2|{|}_{\mathrm{\∞}}\≤\mathrm{\ϵ}and||x-z|{|}_{\mathrm{\∞}}\≤\mathrm{\ϵ}$

Wherein, || || _∞for the ∞ norm of matrix, ε is the second default fault-tolerant value.

As further prioritization scheme of the present invention, use non local self-similarity to carry out further quality enhancing to high-quality test pattern block in step 3, be specially:

For the high-quality test pattern block X of location of pixels (i, j) _h ^p(i, j), finds the image block b similar to it in the high-quality test pattern block of rest of pixels position _n, n={1,2 ..., L}, L are similar piece of number, and to X _h ^p(i, j) carries out following linear expression:

${X_H}^p(i,j)=\underset{{X_H}^p\left(i,j\right)}{\arg \min }{\Σ}_{r=1}^M{x}_r{Y_H}^{rp}(i,j)+\mathrm{\δ}\left|\right|{X_H}^p(i,j)-{\Σ}_{n=1}^L{w}_n{b}_n|{|}_2^2$

Wherein, be the high quality graphic block that the image block of location of pixels (i, j) in r low quality training sample is corresponding, δ is the 3rd regularization parameter, w _nfor linear combination coefficient.

The present invention adopts above technical scheme compared with prior art, there is following technique effect: a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical of the present invention, nuclear norm measurement criterion can be made full use of and portray the structural information be hidden in noise image, obtain the expression coefficient to noise robustness, make the high quality graphic obtained to construct noise more robust, for the process of successive image and identification are offered help.

Accompanying drawing explanation

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

Embodiment

Below in conjunction with the drawings and specific embodiments, technical scheme of the present invention is described in further detail:

The invention provides a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical, idiographic flow as shown in Figure 1.

(1) in image, centered by each location of pixels, the image block of inferior quality test pattern and each location of pixels of training sample image is obtained.

(1) lower quality version of training image is obtained.

Suppose that the size of image is w × h, decimation factor is q, and the image set of training sample composition is expressed as three-dimensional matrice r={1,2, L, M}), wherein, M represents training sample number.First, the low resolution training sample image collection obtained by bicubic interpolation, is expressed as three-dimensional matrice then by the low quality training sample graph image set that bicubic interpolation obtains, be expressed as three-dimensional matrice then,

(2) in image, centered by each location of pixels, the image block of inferior quality test pattern and each location of pixels of training sample image is obtained.

In image (size is for w × h) centered by each location of pixels, get the image block that size is d × d.For inferior quality test pattern its image block set can be expressed as wherein, (i, j) represents image block location of pixels, and N represents the number of image block; For low quality training sample graph image set Y _l, the image block set of its each training sample can be expressed as for high quality training sample graph image set Y _h, the image block set of its each training sample can be expressed as

(2) to each image block in inferior quality test pattern, robust nuclear norm canonical homing method is used to obtain the robust linear expression of its image block set in low quality training sample image on correspondence position.

For inferior quality test pattern X _lin the image block X of location of pixels (i, j) _l ^p(i, j), concentrate the image block of correspondence position to carry out linear expression to it with low quality training sample image, be specially:

X _L ^p(i,j)＝x ₁(i,j)Y _L ^1p(i,j)+x ₂(i,j)Y _L ^2p(i,j)+…+x _M(i,j)Y _L ^Mp(i,j)+e

Wherein, x ₁(i, j), x ₂(i, j) ..., x _m(i, j) represents coefficient, and e is residual error item.

For the purpose of simple and easy, solve in the process representing coefficient following, replace X with D _l ^p(i, j), uses C ₁, C ₂..., C _mreplace Y _l ^1p(i, j), Y _l ^2p(i, j) ..., Y _l ^mp(i, j), uses x ₁, x ₂..., x _mreplace x ₁(i, j), x ₂(i, j) ..., x _m(i, j), then C _rrepresent the image block of location of pixels (i, j) in r low quality training sample, D represents the inferior quality test pattern block of location of pixels (i, j).Further, define from space arrive linear mapping C (x)=x ₁c ₁+ x ₂c ₂+ ... + x _mc _m.

Therefore, the expression coefficient of test pattern block can be obtained by following model solution:

$\underset{x}{\min }\left|\right|x|{|}_q+\mathrm{\λ}||C\left(x\right)-D|{|}_{*}$

Wherein, || x|| _qthe q norm representing coefficient vector x, x _rfor r element value in x; λ is the first regularization parameter; || C (x)-D|| _*it is the nuclear norm all singular values of C (the x)-D (and) of residual error C (x)-D.

In actual applications, due to the impact of the factors such as the motion blur of target image and the noise of equipment self, the image obtained often noise is large, and characteristic details information is also extremely limited.Therefore, the expression coefficient degree of accuracy using above model to try to achieve is not good enough.Therefore, by means of the thought of robust regression, intend adopting and carry out representing solving of coefficient with drag:

$\underset{x}{\min }\left|\right|x|{|}_q+\mathrm{\λ}|\left|Wo\right(C\left(x\right)-D\left)\right|{|}_{*}$

Wherein, W is weight matrix, the i-th row jth column element of W α and β is the parameter pre-set, Y _i,jfor representing the i-th row jth column element value of residual matrix Y; The Hadamard of ο representing matrix amasss.

Represent that coefficient vector x can be solved by following iterative manner:

(1) greatest iteration coefficient is set, when putting primary iteration number of times t=0.

(2) upgrade weight matrix W and represent residual matrix Y, being specially:

$W_{i,j}=\frac{\exp \left(\mathrm{\α}\mathrm{\β}-\mathrm{\α}{\left(Y_{i,j}\right)}^2\right)}{1+xp\left(\mathrm{\α}\mathrm{\β}-\mathrm{\α}{\left(Y_{i,j}\right)}^2\right)}$

Y＝D-D ^(t)

Wherein, D ^(t)the inferior quality test pattern block of location of pixels (i, j) when being the t time iteration, and D ⁽⁰⁾=D _ini, d _iniit is the average image of low quality training sample.

(3) inferior quality test pattern block and low quality training image block is upgraded:

A＝WoD

B _r＝WοC _r

Wherein, A is the inferior quality test pattern block of location of pixels (i, j) after weighting, B _rit is the image block of location of pixels (i, j) in r low quality training sample after weighting.

(4) solving model upgrading robust linear expression coefficient is:

$\underset{x}{\min }\left|\right|x|{|}_q+\mathrm{\λ}||G\left(x\right)-A|{|}_{*},$

Wherein, $G\left(x\right)={\Σ}_{r=1}^M{x}_r{B}_r.$

Above model can be expressed as further:

$\underset{x,z,E}{\min }\left|\right|z|{|}_q+\mathrm{\λ}|\left|E\right|{|}_{*}$

s.t.G(x)-A＝Eandx＝z

Its Lagrangian function is expressed as:

$L_{\mathrm{\μ}}(x,z,E)=\left|\right|z|{|}_q+\mathrm{\λ}|\left|E\right|{|}_{*}+\frac{\mathrm{\μ}}{2}\left(\right||x-z+\frac{1}{\mathrm{\μ}}{y}_1|{|}_F^2+\left|\right|G\left(x\right)-A-E+\frac{1}{\mathrm{\μ}}{Y}_2\left|{|}_F^2\right)-\frac{1}{2\mathrm{\μ}}\left|\right|y_1|{|}_2^2-\frac{1}{2\mathrm{\μ}}|\left|Y_2\right|{|}_F^2.$

Wherein, y ₁, Y ₂be Lagrange multiplier, μ is regularization parameter, for the F norm of matrix.

Adopt alternating direction multiplier method ADMM to solve this model, detailed process is as follows:

A () fixes x, E, upgrade z;

1) as q=1, z can be tried to achieve by soft threshold method:

$z^{l+1}=\underset{z}{\arg \min }\frac{1}{\mathrm{\μ}}\left|\right|z|{|}_1+\frac{1}{2}||z-\left(x^l+\frac{1}{\mathrm{\μ}}{y}_1^l\right)|{|}_F^2$

2) as q=2:

$z^{l+1}=\frac{\mathrm{\μ}}{\mathrm{\μ}+2}\left(x^l+\frac{1}{\mathrm{\μ}}{y}_1^l\right)$

Wherein, z ^l+1be the value of z after l+1 step upgrades, x ^l, y ₁ ^lbe respectively l step and upgrade rear x, y ₁value.

B () fixes z, E, upgrade x:

$x^{l+1}=(H^{T}H+I)\backslash \[H^{T}Vec\left(A+E^l-\frac{1}{\mathrm{\μ}}{Y}_2^l\right)+z^{l+1}-\frac{1}{\mathrm{\μ}}{y}_1^l\];$

Wherein, x ^l+1be the value of x after l+1 step upgrades, H=[Vec (B ₁) ..., Vec (B _m)], the vectorization operation of Vec () representing matrix, E ^l, y ₁ ^l, Y ₂ ^lbe respectively l step and upgrade rear E, y ₁, Y ₂value, I is unit matrix.

C () fixes x, z, upgrade E:

$E^{l+1}=\underset{E}{\arg \min }\frac{\mathrm{\λ}}{\mathrm{\μ}}\left|\right|E|{|}_{*}+\frac{1}{2}||E-(G\left(x^{l+1}\right)-A+\frac{1}{\mathrm{\μ}}{Y}_2^l\left)\right|{|}_F^2;$

Optimum E can be solved by singular value thresholding ^l+1, wherein, E ^l+1be the value of E after l+1 step upgrades, $G\left(x^{l+1}\right)={\Σ}_{r=1}^M{x_r}^{l+1}{B}_r.$

D () upgrades Lagrange multiplier:

$y_1^{l+1}=y_1^l+\mathrm{\μ}(x^{l+1}-z^{l+1})$

$Y_2^{l+1}=Y_2^l+\mathrm{\μ}(G\left(x^{l+1}\right)-A-E^{l+1})$

Wherein, be respectively l+1 step and upgrade rear y ₁, Y ₂value.

If e () reaches following end condition, export x ^l+1as x _r ^(t); Otherwise, turn back to (a):

$\left|\right|G\left(x\right)-A-E+\frac{1}{\mathrm{\μ}}{Y}_2|{|}_{\mathrm{\∞}}\≤\mathrm{\ϵ}and||x-z|{|}_{\mathrm{\∞}}\≤\mathrm{\ϵ}$

Wherein, || || _∞for the ∞ norm of matrix, ε is the second default fault-tolerant value.

(5) inferior quality test pattern block is reconstructed:

$D^{\left(t\right)}={\mathrm{\Σ}}_{r=1}^M{x_r}^{\left(t\right)}{B}_r$

Wherein, D ^(t)inferior quality test pattern block when being the t time iteration;

(6) if reach maximum iteration time or following end condition, export and represent coefficient vector x ^(t)as final solving result; Otherwise, put iterations t=t+1, return step 2.2;

||W ^(t)-W ^(t-1)|| ₂/||W ^(t-1)|| ₂≤γ

Wherein, γ is the first default fault-tolerant value.

(3) when keeping robust representation coefficient constant, low quality training sample image block is replaced with high quality training sample image block, thus obtain high-quality test pattern block corresponding to inferior quality test pattern block, more non local self-similarity is used to carry out further quality enhancing to high-quality test pattern block.

(1) when keeping robust representation coefficient constant, low quality training sample image block is replaced with high quality training sample image block.

The inferior quality test pattern block X of location of pixels (i, j) _l ^pthe high-quality test pattern block of the location of pixels (i, j) that (i, j) is corresponding can be expressed as:

${X_H}^p(i,j)={\Σ}_{r=1}^M{x}_r(i,j){Y_H}^{rp}(i,j)$

Wherein, x _rsolve during (i, j) is (two) obtain robust representation coefficient.

(2) use non local self-similarity, further quality enhancing is carried out to high-quality test pattern block.

For the high-quality test pattern block X of location of pixels (i, j) _h ^p(i, j), finds the image block b similar to it in the high-quality test pattern block of rest of pixels position _n, n={1,2 ..., L}, L are similar piece of number, and to X _h ^p(i, j) carries out following linear expression:

${X_H}^p\left(i,j\right)=\underset{{X_H}^p\left(i,j\right)}{\arg \min }{\Σ}_{r=1}^M{x}_r\left(i,j\right){Y_H}^{rp}\left(i,j\right)+\mathrm{\δ}\left|\right|{X_H}^p\left(i,j\right)-{\Σ}_{n=1}^L{w}_n{b}_n|{|}_2^2$

Wherein, Y _h ^rp(i, j) is _rthe high quality graphic block that in individual low quality training sample, the image block of location of pixels (i, j) is corresponding, δ is the 3rd regularization parameter, w _nfor linear combination coefficient.

(4) the high-quality test pattern block that quality further in (three) strengthens connected and integrated, thus obtaining high-quality test pattern.

The high-quality test pattern block of acquisition is stitched together, thus obtains inferior quality test pattern X _lcorresponding high-quality test pattern X _h.Wherein, for the pixel value of overlapping region, using mean value as last result.

Here, the number of pixels of overlapping region has following two kinds of situations:

(1) if tile size d is odd number, then image block X _h ^pthe coincident pixel number that (i, j) is adjacent block is d × [(d-1)/2];

(2) if tile size d is even number, then image block X _h ^pthe coincident pixel number that (i, j) is adjacent block is d × (d/2).

The above; be only the embodiment in the present invention; but protection scope of the present invention is not limited thereto; any people being familiar with this technology is in the technical scope disclosed by the present invention; the conversion or replacement expected can be understood; all should be encompassed in and of the present inventionly comprise within scope, therefore, protection scope of the present invention should be as the criterion with the protection domain of claims.

Claims (8)

1., based on the wind-powered electricity generation method for enhancing image quality that robust nuclear norm canonical returns, it is characterized in that, comprise following concrete steps:

Step 1, in image centered by each location of pixels, obtains the image block of inferior quality test pattern and each location of pixels of training sample image;

Step 2, to each image block in inferior quality test pattern, uses robust nuclear norm canonical homing method to obtain the robust linear expression of its image block set in low quality training sample image on correspondence position;

Step 3, when keeping robust representation coefficient constant, low quality training sample image block is replaced with high quality training sample image block, thus obtain high-quality test pattern block corresponding to inferior quality test pattern block, more non local self-similarity is used to carry out further quality enhancing to high-quality test pattern block;

Step 4, connects to the high-quality test pattern block that quality further in step 3 strengthens and integrates, thus obtaining high-quality test pattern.

2. a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical according to claim 1, it is characterized in that, centered by each location of pixels, obtain the image block of inferior quality test pattern and each location of pixels of training sample image, be specially in image in step 1:

1.1) lower quality version of training image is obtained;

1.2) in image, centered by each location of pixels, the image block of inferior quality test pattern and each location of pixels of training sample image is obtained.

3. a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical according to claim 2, is characterized in that, by the lower quality version of twice bicubic interpolation acquisition training image in step 1.1.

4. a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical according to claim 1, it is characterized in that, to each image block in inferior quality test pattern in step 2, use robust nuclear norm canonical homing method to obtain the robust linear expression of its image block set in low quality training sample image on correspondence position, be specially:

D＝x ₁C ₁+x ₂C ₂+…+x _MC _M+e

Wherein, D is the inferior quality test pattern block of location of pixels (i, j); C _rthe image block of position (i, j) in r inferior quality pixel training sample, r={1,2 ..., M}, M are training sample numbers; x _rfor representing r element value in coefficient vector x;

Based on robust regression thought, described robust linear expression coefficient vector x obtains according to following model solution:

In formula, || x|| _qthe q norm representing coefficient vector x, λ is the first regularization parameter; W is weight matrix, the i-th row jth column element of W α and β is the parameter pre-set, Y _i,jfor representing the i-th row jth column element value of residual matrix Y; The Hadamard of o representing matrix amasss; C (x)=x ₁c ₁+ x ₂c ₂+ ... + x _mc _mfrom space arrive a linear mapping, d × d is the size of image block; || || _*represent nuclear norm.

5. a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical according to claim 4, is characterized in that, according to model the method solving robust linear expression coefficient vector x is as follows:

2.1) set greatest iteration coefficient, put primary iteration number of times t=0;

2.2) upgrade weight matrix W and represent residual matrix Y, being specially:

Y＝D-D ^(t)

Wherein, D ^(t)the inferior quality test pattern block of location of pixels (i, j) when being the t time iteration, and D ⁽⁰⁾=D _ini, D _inithe average image of low quality training sample,

2.3) upgrade inferior quality test pattern block and low quality training image block, be specially:

A＝WоD

B _r＝WоC _r

Wherein, A is the inferior quality test pattern block of location of pixels (i, j) after weighting, B _rit is the image block of location of pixels (i, j) in r low quality training sample after weighting; 2.4) Renewal model be specially:

s.t.G(x)-A＝Eandx＝z

Its Lagrangian function is expressed as:

Wherein, y ₁, Y ₂be Lagrange multiplier, μ is the second regularization parameter, for the F norm of matrix;

2.5) alternating direction multiplier method ADMM is adopted to solve the model after renewal in step 2.5, the expression coefficient x exported when obtaining the t time iteration _r ^(t), thus the expression coefficient vector x exported when obtaining the t time iteration ^(t);

2.6) reconstruct inferior quality test pattern block, be specially:

Wherein, D ^(t)inferior quality test pattern block when being the t time iteration;

2.7) if reach maximum iteration time or following end condition, export and represent coefficient vector x ^(t)as final solving result; Otherwise, put t=t+1, return step 2.2;

||W ^(t)-W ^(t-1)|| ₂/||W ^(t-1)|| ₂≤γ

Wherein, γ is the first default fault-tolerant value.

6. a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical according to claim 5, it is characterized in that, in step 2.2, maximum iteration time is more than or equal to 2.

7. a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical according to claim 5, is characterized in that, adopt alternating direction multiplier method ADMM to solve the model in step 2.5, be specially in step 2.5:

2.5.1) fixing x, E, upgrades z, is specially:

A) as q=1, z is tried to achieve by soft threshold method:

B) as q=2,

Wherein, z ^l+1be the value of z after l+1 step upgrades, x ^l, y ₁ ^lbe respectively l step and upgrade rear x, y ₁value;

2.5.2) fixing z, E, upgrades x, is specially:

Wherein, x ^l+1be the value of x after l+1 step upgrades, H=[Vec (B ₁) ..., Vec (B _m)], the vectorization operation of Vec () representing matrix, E ^l, y ₁ ^l, Y ₂ ^lbe respectively l step and upgrade rear E, y ₁, Y ₂value;

2.5.3) fixing x, z, upgrades E, is specially:

Optimum E is solved by singular value thresholding ^l+1:

Wherein, E ^l+1be the value of E after l+1 step upgrades,

2.5.4) upgrade Lagrange multiplier, be specially:

Wherein, be respectively l+1 step and upgrade rear y ₁, Y ₂value;

2.5.5) if reach following end condition, then x is exported ^l+1as otherwise, turn back to step 2.5.1;

Wherein, || || _∞for the ∞ norm of matrix, ε is the second default fault-tolerant value.

8. a kind of wind-powered electricity generation method for enhancing image quality returned based on robust nuclear norm canonical according to claim 5, is characterized in that, use non local self-similarity to carry out further quality enhancing to high-quality test pattern block, be specially in step 3:

For the high-quality test pattern block X of location of pixels (i, j) _h ^p(i, j), finds the image block b similar to it in the high-quality test pattern block of rest of pixels position _n, n={1,2 ..., L}, L are similar piece of number, and to X _h ^p(i, j) carries out following linear expression:

Wherein, Y _h ^rp(i, j) is _rthe high quality graphic block that in individual low quality training sample, the image block of location of pixels (i, j) is corresponding, δ is the 3rd regularization parameter, w _nfor linear combination coefficient.