CN105761287A - Image decomposition method and device based on sparse representation - Google Patents

Abstract

The invention provides an image decomposition method and device based on sparse representation. The method comprises the steps: enabling a to-be-decomposed image to be decomposed into a plurality of image pieces; obtaining a cartoon image piece corresponding to each image piece according to an image sparse representation model, the plurality of image pieces and a pre-trained cartoon dictionary, and obtaining a texture image piece corresponding to each image piece according to the image sparse representation model, the plurality of image pieces and a pre-trained texture dictionary; obtaining a cartoon image corresponding to the to-be-decomposed image according to each cartoon image piece, and obtaining the texture image corresponding to the to-be-decomposed image according to each texture image piece. The method introduces the sparse representation into image decomposition, depicts components of the image at different frequencies through the pre-obtained cartoon dictionary and the texture dictionary, is high in adaptive degree, and is small in number of stripe residuals after image decomposition. In addition, the method also can be used for the training of the cartoon dictionary and the corresponding dual dictionary, can speed up the image decomposition through the introduction of the dual dictionary, and improves the efficiency of image decomposition.

Description

A kind of picture breakdown method based on rarefaction representation and device

Technical field

The present invention relates to technical field of image processing, in particular to a kind of picture breakdown method based on rarefaction representation and device.

Background technology

Generally comprising stripe information in image, stripe information corresponds to the high-frequency information in image, typically requires the stripe information removed in image in image processing process, and image goes the process of striped to be the catabolic process of image.

Picture breakdown process is to be cartoon composition and texture composition by picture breakdown.Correlation technique generally adopt filter equalizer method carry out exploded view picture, adopt the wave filter of a fixed size, according to preset order, image is filtered, function space by the cartoon Partial Transformation of image to bounded variation, texture part is transformed in texture space, thus realizing the cartoon composition of image and the decomposition of texture composition.

But the adaptive ability of correlation technique median filter is very poor, it is necessary to by being manually adjusted, cause that the efficiency of picture breakdown is very low, decompose in the cartoon part obtained and remain a lot of stripe information.

Summary of the invention

In view of this, the purpose of the embodiment of the present invention is in that to provide a kind of picture breakdown method based on rarefaction representation and device, rarefaction representation is introduced in picture breakdown, different frequency composition by the cartoon dictionary of training in advance and texture dictionary picture engraving, self adaptation degree is high, striped little residue after picture breakdown.

First aspect, embodiments provides a kind of picture breakdown method based on rarefaction representation, and described method includes:

Treat exploded view picture and carry out burst process, obtain multiple image sheet；

The cartoon dictionary of model, the plurality of image sheet and training in advance is represented according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding respectively, represent the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtain the texture maps photo that each image sheet is corresponding respectively；

According to the cartoon image sheet that described each image sheet is corresponding, obtain the cartoon image that described image to be decomposed is corresponding, and according to texture maps photo corresponding to described each image sheet, obtain the texture image that described image to be decomposed is corresponding.

In conjunction with first aspect, embodiments provide the first possible implementation of above-mentioned first aspect, wherein, the described cartoon dictionary representing model, the plurality of image sheet and training in advance according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding respectively, including:

Represent the cartoon dictionary of model, the plurality of image sheet and training in advance according to image sparse, obtain the cartoon sparse coefficient that each image sheet is corresponding respectively；

According to the cartoon sparse coefficient that described cartoon dictionary is corresponding with described each image sheet, calculate the cartoon image sheet that described each image sheet is corresponding respectively.

In conjunction with first aspect, embodiments provide the implementation that the second of above-mentioned first aspect is possible, wherein, the described cartoon image sheet corresponding according to described each image sheet, obtain the cartoon image that described image to be decomposed is corresponding, including:

According to described each image sheet position in described image to be decomposed, cartoon image sheet corresponding for described each image sheet is combined as complete cartoon image；

From described complete cartoon image, remove overlapping region, obtain the cartoon image that described image to be decomposed is corresponding.

In conjunction with first aspect, embodiments provide the third possible implementation of above-mentioned first aspect, wherein, the described texture dictionary representing model, the plurality of image sheet and training in advance according to described image sparse, obtain the texture maps photo that each image sheet is corresponding respectively, including:

Represent the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtain the texture sparse coefficient that each image sheet is corresponding respectively；

According to the texture sparse coefficient that described texture dictionary is corresponding with described each image sheet, calculate the texture maps photo that described each image sheet is corresponding respectively.

In conjunction with first aspect, embodiments provide the 4th kind of possible implementation of above-mentioned first aspect, wherein, the described texture maps photo corresponding according to described each image sheet, obtain the texture image that described image to be decomposed is corresponding, including:

According to described each image sheet position in described image to be decomposed, texture maps photo corresponding for described each image sheet is combined as complete texture image；

From described complete texture image, remove overlapping region, obtain the texture image that described image to be decomposed is corresponding.

Second aspect, embodiments provides a kind of picture breakdown method based on rarefaction representation, and described method includes:

Treat exploded view picture and carry out burst process, obtain multiple image sheet；

Represent, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon image that described image to be decomposed is corresponding；

From described image to be decomposed, deduct described cartoon image, obtain the texture image that described image to be decomposed is corresponding.

In conjunction with second aspect, embodiments provide the first possible implementation of above-mentioned second aspect, wherein, described represent the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance according to image sparse, obtain the cartoon image that described image to be decomposed is corresponding, including:

Represent, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon sparse coefficient that each image sheet is corresponding respectively；

The cartoon sparse coefficient that cartoon dictionary according to training in advance is corresponding with described each image sheet, obtains the cartoon image that described image to be decomposed is corresponding.

The first possible implementation in conjunction with second aspect, embodiments provide the implementation that the second of above-mentioned second aspect is possible, wherein, the described cartoon sparse coefficient corresponding with described each image sheet according to the cartoon dictionary of training in advance, obtain the cartoon image that described image to be decomposed is corresponding, including:

The cartoon sparse coefficient that cartoon dictionary according to training in advance is corresponding with described each image sheet, calculates the cartoon image sheet that described each image sheet is corresponding respectively；

According to described each image sheet position in described image to be decomposed, cartoon image sheet corresponding for described each image sheet is combined as complete cartoon image；

From described complete cartoon image, remove overlapping region, obtain the cartoon image that described image to be decomposed is corresponding.

The third aspect, embodiments provides a kind of picture breakdown device based on rarefaction representation, and described device includes:

Burst module, is used for treating exploded view picture and carries out burst process, obtain multiple image sheet；

First acquisition module, for representing the cartoon dictionary of model, the plurality of image sheet and training in advance according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding respectively, represent the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtain the texture maps photo that each image sheet is corresponding respectively；

Second acquisition module, for according to cartoon image sheet corresponding to described each image sheet, obtains the cartoon image that described image to be decomposed is corresponding, and according to texture maps photo corresponding to described each image sheet, obtains the texture image that described image to be decomposed is corresponding.

In conjunction with the third aspect, embodiments providing the first possible implementation of the above-mentioned third aspect, wherein, described first acquisition module includes:

First acquiring unit, for representing the cartoon dictionary of model, the plurality of image sheet and training in advance according to image sparse, obtains the cartoon sparse coefficient that each image sheet is corresponding respectively；

First computing unit, for the cartoon sparse coefficient corresponding with described each image sheet according to described cartoon dictionary, calculates the cartoon image sheet that described each image sheet is corresponding respectively.

In conjunction with the third aspect, embodiments providing the implementation that the second of the above-mentioned third aspect is possible, wherein, described second acquisition module includes:

First assembled unit, for according to described each image sheet position in described image to be decomposed, being combined as complete cartoon image by cartoon image sheet corresponding for described each image sheet；

First removal unit, for removing overlapping region from described complete cartoon image, obtains the cartoon image that described image to be decomposed is corresponding.

In conjunction with the third aspect, embodiments providing the third possible implementation of the above-mentioned third aspect, wherein, described first acquisition module includes:

Second acquisition unit, for representing the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtains the texture sparse coefficient that each image sheet is corresponding respectively；

Second computing unit, for the texture sparse coefficient corresponding with described each image sheet according to described texture dictionary, calculates the texture maps photo that described each image sheet is corresponding respectively.

In conjunction with the third aspect, embodiments providing the 4th kind of possible implementation of the above-mentioned third aspect, wherein, described second acquisition module includes:

Second assembled unit, for according to described each image sheet position in described image to be decomposed, being combined as complete texture image by texture maps photo corresponding for described each image sheet；

Second removal unit, for removing overlapping region from described complete texture image, obtains the texture image that described image to be decomposed is corresponding.

Fourth aspect, embodiments provides a kind of picture breakdown device based on rarefaction representation, and described device includes:

Burst module, is used for treating exploded view picture and carries out burst process, obtain multiple image sheet；

Acquisition module, for representing, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtains the cartoon image that described image to be decomposed is corresponding；

Deduct module, for deducting described cartoon image from described image to be decomposed, obtain the texture image that described image to be decomposed is corresponding.

In conjunction with fourth aspect, embodiments providing the first possible implementation of above-mentioned fourth aspect, wherein, described acquisition module includes:

First acquiring unit, for representing, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtains the cartoon sparse coefficient that each image sheet is corresponding respectively；

Second acquisition unit, for the cartoon sparse coefficient corresponding with described each image sheet according to the cartoon dictionary of training in advance, obtains the cartoon image that described image to be decomposed is corresponding.

In conjunction with the first possible implementation of fourth aspect, embodiments providing the implementation that the second of above-mentioned fourth aspect is possible, wherein, described second acquisition unit includes:

Computation subunit, for the cartoon sparse coefficient corresponding with described each image sheet according to the cartoon dictionary of training in advance, calculates the cartoon image sheet that described each image sheet is corresponding respectively；

Assembled unit, for according to described each image sheet position in described image to be decomposed, being combined as complete cartoon image by cartoon image sheet corresponding for described each image sheet；

Removal unit, for removing overlapping region from described complete cartoon image, obtains the cartoon image that described image to be decomposed is corresponding.

In the method and device of embodiment of the present invention offer, image slices to be decomposed is become multiple image sheet；The cartoon dictionary of model, multiple image sheet and training in advance is represented according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding, represent the texture dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the texture maps photo that each image sheet is corresponding；According to each cartoon image sheet, obtain the cartoon image that image to be decomposed is corresponding, and according to each texture maps photo, obtain the texture image that image to be decomposed is corresponding.Rarefaction representation is introduced in picture breakdown by the present invention, and by the different frequency composition of the cartoon dictionary of training in advance and texture dictionary picture engraving, self adaptation degree is high, striped little residue after picture breakdown.Furthermore it is also possible to only train the antithesis dictionary of cartoon dictionary and correspondence thereof, introduce antithesis dictionary and can accelerate picture breakdown speed, improve picture breakdown efficiency.

For making the above-mentioned purpose of the present invention, feature and advantage to become apparent, preferred embodiment cited below particularly, and coordinate appended accompanying drawing, it is described in detail below.

Accompanying drawing explanation

In order to be illustrated more clearly that the technical scheme of the embodiment of the present invention, the accompanying drawing used required in embodiment will be briefly described below, it is to be understood that, the following drawings illustrate only certain embodiments of the present invention, therefore the restriction to scope it is not construed as, for those of ordinary skill in the art, under the premise not paying creative work, it is also possible to obtain other relevant accompanying drawings according to these accompanying drawings.

Fig. 1 illustrates the flow chart of a kind of picture breakdown method based on rarefaction representation that the embodiment of the present invention 1 provides；

Fig. 2 illustrates the flow chart of a kind of picture breakdown method based on rarefaction representation that the embodiment of the present invention 2 provides；

Fig. 3 illustrates the structural representation of a kind of picture breakdown device based on rarefaction representation that the embodiment of the present invention 3 provides；

Fig. 4 illustrates the structural representation of a kind of picture breakdown device based on rarefaction representation that the embodiment of the present invention 4 provides.

Detailed description of the invention

Below in conjunction with accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only a part of embodiment of the present invention, rather than whole embodiments.Generally can with various different configurations arrange and design with the assembly of the embodiment of the present invention that illustrate described in accompanying drawing herein.Therefore, below the detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit claimed the scope of the present invention, but is merely representative of the selected embodiment of the present invention.Based on embodiments of the invention, the every other embodiment that those skilled in the art obtain under the premise not making creative work, broadly fall into the scope of protection of the invention.

Considering to adopt filter equalizer method to carry out exploded view picture in correlation technique, the adaptive ability of wave filter is very poor, it is necessary to by being manually adjusted, and causes that the efficiency of picture breakdown is very low, decomposes in the cartoon part obtained and remains a lot of stripe information.Based on this, embodiments provide a kind of picture breakdown method based on rarefaction representation and device.It is described by the examples below.

Embodiment 1

Embodiments provide a kind of picture breakdown method based on rarefaction representation.

The basic norm of rarefaction representation is by transmitting the least possible information to characterize more information.Rarefaction representation is introduced in picture breakdown by the embodiment of the present invention, by the mode training in advance cartoon dictionary of sparse coding and texture dictionary.Sparse coding is one finds out the problem that signal correspondence crosses the rarefaction representation of complete dictionary.

In embodiments of the present invention, represented complete dictionary with letter D, and crossed complete dictionary D and generally obtain from training sample set learning.Training sample set includes multiple image, and this training sample set is represented by X={x₁,x₂,...,x_n, for each image in training sample set, respectively by image sheet that each image burst is multiple pre-set dimension.When training complete dictionary D, the image sheet of multiple pre-set dimension corresponding to each image is trained study and can obtain complete dictionary D.The expression formula crossing complete dictionary D is as follows:

$\begin{array}{cc}D=\arg \underset{D,Z}{\min }\left|\right|X-DZ|{|}_2^2+\mathrm{\λ}|\left|Z\right|{|}_1& s.t.\left|\right|D_i|{|}_2^2\≤1,i=1,2,\mathrm{...},K\end{array}\mathrm{...}\left(1\right)$

Wherein, in formula (1), D represented complete dictionary, and Z represents the sparse coefficient of the image sheet of corresponding pre-set dimension, and i represents the sequence number of image sheet in the training set comprising K image sheet, and λ is the sparse penalty term factor,Obscure for removing yardstick,Norm | | Z | |₁For implementing degree of rarefication.

According to above-mentioned formula (1) iteration optimization D and Z.The optimization process solving D and Z is:

1) adopt gaussian random matrix initialisation to cross complete dictionary D, and the every string crossing complete dictionary D is normalized；

2) fix complete dictionary D, update Z by equation below (2),

$Z=\arg \underset{Z}{\min }\left|\right|X-DZ|{|}_2^2+\mathrm{\λ}|\left|Z\right|{|}_1\mathrm{...}\left(2\right)$

In the embodiment of the present invention, can quickly solve Z by linear program.

3) fixing Z, by the updated complete dictionary D of equation below (3).

$\begin{array}{cc}D=\arg \underset{D}{\min }\left|\right|X-DZ|{|}_2^2& s.t.\left|\right|D_i|{|}_2^2\≤1,i=1,2,\mathrm{...},K\end{array}\mathrm{...}\left(3\right)$

In embodiments of the present invention, the training method of cartoon dictionary and texture dictionary is identical with the training method of above-mentioned mistake complete dictionary D.For cartoon image and texture image, by above-mentioned steps 1)-3) solve optimal solution respectively, it is possible to obtain corresponding cartoon dictionary D_carWith texture dictionary D_tex.Cartoon dictionary D_carFor the combination of multiple cartoon image sheets, texture dictionary D_texCombination for multiple texture maps photos.The size of the image sheet in dictionary is also above-mentioned pre-set dimension.

Referring to Fig. 1, training obtains cartoon dictionary D by the way_carWith texture dictionary D_texAfter, decompose image to be decomposed by the operation of step 101-103.

Step 101: treat exploded view picture and carry out burst process, obtain multiple image sheet.

According to preset order in image to be decomposed, image slices to be decomposed is become the image sheet of multiple pre-set dimension.Preset order can be start first from left to right order from top to bottom again from the upper left corner, or for starting first order etc. from top to bottom more from right to left from the upper right corner, the embodiment of the present invention does not specifically limit particular order during burst, practical operation can determine according to demand adopt which kind of order to carry out burst.What adopt when the image in training sample set being carried out burst in above-mentioned pre-set dimension and dictionary training process is equivalently-sized, and pre-set dimension can be 3 × 3,5 × 5 or 7 × 7 etc., and 3,5 and 7 represent number of pixels.In embodiments of the present invention, representing image sheet with symbol y, being shaped as of image sheet is square.

Step 102: represent the cartoon dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding respectively, represent the texture dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the texture maps photo that each image sheet is corresponding respectively.

In embodiments of the present invention, it is assumed that signal b was expressed as the sparse linear combination of complete dictionary D, then represent model according to image sparse, and signal b can be written as b=D α₀, α₀It is a vector and there is little nonzero term.

Then each image sheet y, y=D α, α for image to be decomposed represent the vector that the image sheet y rarefaction representation coefficient relative to dictionary D forms, it is possible to solve the image sheet y rarefaction representation coefficient relative to dictionary D by equation below (4),

$\begin{array}{ccc}min\left|\right|\mathrm{\α}|{|}_0& s.t.& \left|\right|FD\mathrm{\α}-Fy|{|}_2^2\≤\mathrm{\ϵ}\end{array}\mathrm{...}\left(4\right)$

Wherein, in formula (4), | | α | |₀Representing zero norm of rarefaction representation coefficient, zero norm represents the number of nonzero element in the vector asking rarefaction representation coefficient to form, and F is Linear feature extraction operator.

Cartoon part and texture part is included due to image, so when solving image sheet that image includes relative to the optimum sparse coefficient of cartoon dictionary or texture dictionary, it is necessary to above-mentioned formula (4) is deformed into formulaWhen solving cartoon sparse coefficient, the dictionary D in deformation formula is replaced with cartoon dictionary D_car, when solving texture sparse coefficient, the dictionary D in above-mentioned deformation formula is replaced with texture dictionary D_tex。

In the embodiment of the present invention, carry out in the process decomposed treating exploded view picture, it is necessary to solve cartoon image sheet corresponding to each image sheet that image to be decomposed includes and texture maps photo respectively.The cartoon image sheet that each image sheet that image to be decomposed includes is corresponding can be solved in the following way, specifically include:

Represent the cartoon dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the cartoon sparse coefficient that each image sheet is corresponding respectively；According to the cartoon sparse coefficient that cartoon dictionary is corresponding with each image sheet, calculate the cartoon image sheet that each image sheet is corresponding respectively.

For any image sheet that image to be decomposed includes, represent the cartoon dictionary of model, this image sheet and training in advance according to image sparse, calculate, by equation below (5), the cartoon sparse coefficient that this image sheet is corresponding.

$min\mathrm{\λ}\left|\right|{\mathrm{\α}}_1|{|}_1+\frac{1}{2}||D_{car}{\mathrm{\α}}_1-y|{|}_2^2\mathrm{...}\left(5\right)$

Wherein, in formula (5), y represents an image sheet of image to be decomposed, α₁Represent cartoon sparse coefficient corresponding to image sheet y, D_carRepresent the cartoon dictionary of training in advance.

After obtaining, by above-mentioned calculating, the cartoon sparse coefficient that this image sheet is corresponding, calculate the product of cartoon sparse coefficient corresponding to this image sheet and the cartoon dictionary of training in advance, obtain the cartoon image sheet that this image sheet is corresponding.

For other each image sheets that image to be decomposed is corresponding, with this image sheet, the cartoon image sheet that other each image sheets are corresponding can be obtained in the manner described above respectively.

Similarly, the embodiment of the present invention solves the texture maps photo that each image sheet that image to be decomposed includes is corresponding in the following way, specifically includes:

Represent the texture dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the texture sparse coefficient that each image sheet is corresponding respectively；According to the texture sparse coefficient that texture dictionary is corresponding with each image sheet, calculate the texture maps photo that each image sheet is corresponding respectively.

For any image sheet that image to be decomposed includes, represent the texture dictionary of model, this image sheet and training in advance according to image sparse, calculate, by equation below (6), the texture sparse coefficient that this image sheet is corresponding.

$min\mathrm{\λ}\left|\right|{\mathrm{\α}}_2|{|}_1+\frac{1}{2}||D_{tex}{\mathrm{\α}}_2-y|{|}_2^2\mathrm{...}\left(6\right)$

Wherein, in formula (5), y represents an image sheet of image to be decomposed, α₂Represent texture sparse coefficient corresponding to image sheet y, D_texRepresent the texture dictionary of training in advance.

After obtaining, by above-mentioned calculating, the texture sparse coefficient that this image sheet is corresponding, calculate the product of texture sparse coefficient corresponding to this image sheet and the texture dictionary of training in advance, obtain the texture maps photo that this image sheet is corresponding.

For other each image sheets that image to be decomposed is corresponding, with this image sheet, the texture maps photo that other each image sheets are corresponding can be obtained in the manner described above respectively.

Step 103: according to the cartoon image sheet that each image sheet is corresponding, obtains the cartoon image that image to be decomposed is corresponding, and according to texture maps photo corresponding to each image sheet, obtains the texture image that image to be decomposed is corresponding.

In the embodiment of the present invention, treat in a step 101 exploded view picture carry out burst process time, also record each image sheet position in image to be decomposed.After obtaining cartoon image sheet corresponding to each image sheet, according to each image sheet position in image to be decomposed, cartoon image sheet corresponding for each image sheet is combined as complete cartoon image.Owing to would be likely to occur the pixel region of overlap between multiple image sheets that image to be decomposed includes, also overlapping region it is inevitably present so combining in the complete cartoon image obtained, therefore must remove overlapping region from complete cartoon image, obtain the cartoon image that image to be decomposed is corresponding.

Similarly, according to each image sheet position in image to be decomposed, texture maps photo corresponding for each image sheet is combined as complete texture image.From this complete texture image, remove overlapping region, obtain the texture image that image to be decomposed is corresponding.

In embodiments of the present invention, image slices to be decomposed is become multiple image sheet；The cartoon dictionary of model, multiple image sheet and training in advance is represented according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding, represent the texture dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the texture maps photo that each image sheet is corresponding；According to each cartoon image sheet, obtain the cartoon image that image to be decomposed is corresponding, and according to each texture maps photo, obtain the texture image that image to be decomposed is corresponding.Rarefaction representation is introduced in picture breakdown by the present invention, and by the different frequency composition of the cartoon dictionary of training in advance and texture dictionary picture engraving, self adaptation degree is high, striped little residue after picture breakdown.

Embodiment 2

Embodiments provide a kind of picture breakdown method based on rarefaction representation.The basic norm of rarefaction representation is by transmitting the least possible information to characterize more information.Rarefaction representation is introduced in picture breakdown by the embodiment of the present invention, owing to the arithmetic speed of sparse learning algorithm is very slow, therefore only training in advance cartoon dictionary in the embodiment of the present invention, and obtain the antithesis dictionary that this cartoon dictionary is corresponding, antithesis dictionary is introduced in picture breakdown, improve the speed of picture breakdown.

In embodiments of the present invention, it is necessary first to train cartoon dictionary, and obtain the antithesis dictionary that this cartoon dictionary is corresponding.Represent image to be decomposed with X, then image X to be decomposed can be expressed as X=X_car+X_tex.Wherein, X_carRepresent cartoon image, X_texRepresent texture image.PADDLE algorithm is incorporated in picture breakdown by the embodiment of the present invention, and the image fast decoupled model based on PADDLE algorithm is:

$E(P_{car},C,U)=\frac{1}{d}\left|\right|X_C-D_{car}U|{|}_F^2+\frac{\mathrm{\η}}{K}||U-CX|{|}_F^2+\frac{2\mathrm{\λ}}{K}\underset{i=1}{\overset{N}{\Σ}}\left|\right|u_i|{|}_1,|\left|d_i\right|{|}^2\≤1,\left|\right|C_i|{|}^2\≤\mathrm{1...}\left(1\right)$

Wherein, in formula (1), E represents energy function；F represents F norm, uses 2 norms in the embodiment of the present invention；X=[x₁,x₂,...,x_N] represent image, x₁,x₂,...,x_NFor image sheet image slices obtained according to pre-set dimension；D_car=[d₁,d₂,...,d_K] representing cartoon dictionary, d represents the size of training image sheet；C=[c₁,c₂,...,c_K] represent cartoon dictionary D_carCorresponding antithesis dictionary；U=[u₁,u₂,...,u_K] for training data coding matrix, u represents the training data of every string in U；I represents the sequence number at training image sheet；λ represents the sparse penalty term factor, and η represents initial code Error weight.

The cartoon dictionary D obtained can be trained by above-mentioned model_carWith its antithesis dictionary C, obtain cartoon dictionary D in training_carBefore its antithesis dictionary C, need to first solving U, the embodiment of the present invention is at fixing cartoon dictionary D_carWith under the premise of its antithesis dictionary C, solve U by equation below (2)-(5):

E (U)=F (U)+J (U) ... (2)

Wherein, in formula (2),Target functionLaunch to obtain by F (U),

$F\left(U\right)=\frac{1}{d}{(X-D_{car}U)}^T(X-D_{car}U)+\frac{\mathrm{\η}}{K}{(U-CX)}^T(U-CX)\mathrm{...}\left(3\right)$

Above-mentioned formula (3) is continued expansion can obtain:

$F\left(U\right)=\frac{1}{d}(X^{T}X-X^T{D}_{car}U-U^T{D}_{car}^{T}X+U^T{D}_{car}^T{D}_{car}U)+\frac{\mathrm{\η}}{K}(U^{T}U-U^{T}CX-X^T{C}^{T}U+X^T{C}^{T}CX)\mathrm{...}\left(4\right)$

According to above-mentioned formula (4), deformation obtains the more new-standard cement of U and is:

$U^p=S_{\mathrm{\λ}/{\mathrm{K\σ}}_U}\[(1-\frac{\mathrm{\η}}{{\mathrm{K\σ}}_U})U^{p-1}+\frac{1}{{\mathrm{\σ}}_U}\left(\frac{1}{d}{D}_{car}^T\right(X-D_{car}{U}^{p-1})+\frac{\mathrm{\η}}{K}CX)\]\mathrm{...}\left(5\right)$

Solve by the way after obtaining U, solve cartoon dictionary D_carDecoupling problem is translated into, it is possible to by fixing cartoon dictionary D with the optimization problem of its antithesis dictionary C_carSolve antithesis dictionary C, or solve cartoon dictionary D by fixing antithesis dictionary C_car。

In above-mentioned formula (2), for target function J, cartoon dictionary D_carRow and capable in its antithesis dictionary C on restriction be the same.With B representation unit ball, target function J is to cartoon dictionary D_carRestriction with its antithesis dictionary C can be described as: cartoon dictionary D_carRow and the row of antithesis dictionary C belong to B.Because J is a target function, so for cartoon dictionary D_carWith antithesis dictionary C, arest neighbors operator is all mapping operator.With π (d)=d/max{1, | | d} represents the mapping on unit ball B, uses π_DRepresent that application π is cartoon dictionary D_carIn row be mapped on unit ball B, use π_CRepresent that application π is mapped to the row in antithesis dictionary C on unit ball B.Fix cartoon dictionary D again_carWith a dictionary in antithesis dictionary C, (4) formula is asked respectively for cartoon dictionary D_carWith the gradient of antithesis dictionary C, combine respective map operation operator π_DAnd π_C, it is possible to obtain cartoon dictionary D_carMore new-standard cement with its antithesis dictionary C:

$D_{car}^p={\mathrm{\π}}_D(D_{car}^{p-1}+\frac{1}{{\mathrm{d\σ}}_{D_{car}}}(X-D_{car}^{p-1}U\left)U^T\right)\mathrm{...}\left(6\right)$

$C^p={\mathrm{\π}}_C(C^{p-1}+\frac{1}{{\mathrm{d\σ}}_C}(X-C^{p-1}X\left)X^T\right)\mathrm{...}\left(7\right)$

Referring to Fig. 2, training obtains cartoon dictionary D by the way_carAfter its antithesis dictionary C, image to be decomposed is decomposed in the operation of 201-203 as follows.

Step 201: treat exploded view picture and carry out burst process, obtain multiple image sheet.

According to preset order in image to be decomposed, image slices to be decomposed is become the image sheet of multiple pre-set dimension.Preset order can be start first from left to right order from top to bottom again from the upper left corner, or for starting first order etc. from top to bottom more from right to left from the upper right corner, the embodiment of the present invention does not specifically limit particular order during burst, practical operation can determine according to demand adopt which kind of order to carry out burst.What adopt when image being carried out burst in above-mentioned pre-set dimension and dictionary training process is equivalently-sized, and pre-set dimension can be 3 × 3,5 × 5 or 7 × 7 etc., and 3,5 and 7 represent number of pixels.In embodiments of the present invention, representing image sheet with symbol y, being shaped as of image sheet is square.

Step 202: represent, according to image sparse, the antithesis dictionary that model, multiple image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon image that image to be decomposed is corresponding.

Represent, according to image sparse, the antithesis dictionary that model, multiple image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon sparse coefficient that each image sheet is corresponding respectively；The cartoon sparse coefficient that cartoon dictionary according to training in advance is corresponding with each image sheet, obtains the cartoon image that image to be decomposed is corresponding.

For any image sheet that image to be decomposed includes, this image sheet and antithesis dictionary are carried out linear multiplication, obtains the cartoon sparse coefficient that this image sheet is corresponding.For other each image sheets that image to be decomposed includes, the cartoon sparse coefficient that other each image sheets are corresponding can be calculated in the manner described above respectively.

Then according to the cartoon sparse coefficient that the cartoon dictionary of training in advance is corresponding with each image sheet, the cartoon image sheet that each image sheet is corresponding is calculated respectively；According to each image sheet position in image to be decomposed, cartoon image sheet corresponding for each image sheet is combined as complete cartoon image；From complete cartoon image, remove overlapping region, obtain the cartoon image that image to be decomposed is corresponding.

In embodiments of the present invention, for each image sheet, calculate the product between cartoon sparse coefficient corresponding to each image sheet and the cartoon dictionary of training in advance respectively, obtain the cartoon image sheet that each image sheet is corresponding.

In the embodiment of the present invention, treat in a step 101 exploded view picture carry out burst process time, also record each image sheet position in image to be decomposed.After obtaining cartoon image sheet corresponding to each image sheet, according to each image sheet position in image to be decomposed, cartoon image sheet corresponding for each image sheet is combined as complete cartoon image.Owing to would be likely to occur the pixel region of overlap between multiple image sheets that image to be decomposed includes, also overlapping region it is inevitably present so combining in the complete cartoon image obtained, therefore must remove overlapping region from complete cartoon image, obtain the cartoon image that image to be decomposed is corresponding.

Step 203: deduct cartoon image from image to be decomposed, obtains the texture image that image to be decomposed is corresponding.

Form owing to image to be decomposed is superposed with texture image by cartoon image, so after obtaining the cartoon image that image to be decomposed is corresponding, directly deducting this cartoon image from image to be decomposed, the texture image that image to be decomposed is corresponding can be obtained.

In embodiments of the present invention, image slices to be decomposed is become multiple image sheet；Represent, according to image sparse, the antithesis dictionary that model, multiple image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon image that image to be decomposed is corresponding；From image to be decomposed, deduct cartoon image, obtain the texture image that image to be decomposed is corresponding.Rarefaction representation is introduced in picture breakdown by the present invention, and by the different frequency composition of the antithesis dictionary picture engraving of the cartoon dictionary of training in advance and correspondence thereof, self adaptation degree is high, striped little residue after picture breakdown.And introduce antithesis dictionary and can accelerate picture breakdown speed, improve picture breakdown efficiency.

Embodiment 3

Referring to Fig. 3, embodiments providing a kind of picture breakdown device based on rarefaction representation, this device is for performing the picture breakdown method based on rarefaction representation that above-described embodiment 1 provides.This device specifically includes:

Burst module 301, is used for treating exploded view picture and carries out burst process, obtain multiple image sheet；

First acquisition module 302, for representing the cartoon dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding respectively, represent the texture dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the texture maps photo that each image sheet is corresponding respectively；

Second acquisition module 303, for according to cartoon image sheet corresponding to each image sheet, obtains the cartoon image that image to be decomposed is corresponding, and according to texture maps photo corresponding to each image sheet, obtains the texture image that image to be decomposed is corresponding.

Above-mentioned first acquisition module 302 obtains, by the first acquiring unit and the first computing unit, the cartoon image sheet that each image sheet is corresponding.

First acquiring unit, for representing the cartoon dictionary of model, multiple image sheet and training in advance according to image sparse, obtains the cartoon sparse coefficient that each image sheet is corresponding respectively；

First computing unit, for the cartoon sparse coefficient corresponding with each image sheet according to cartoon dictionary, calculates the cartoon image sheet that each image sheet is corresponding respectively.

Above-mentioned second acquisition module 303 obtains, by the first assembled unit and the first removal unit, the cartoon image that image to be decomposed is corresponding.

First assembled unit, for according to each image sheet position in image to be decomposed, being combined as complete cartoon image by cartoon image sheet corresponding for each image sheet；

First removal unit, for removing overlapping region from complete cartoon image, obtains the cartoon image that image to be decomposed is corresponding.

Above-mentioned first acquisition module 302 obtains, by second acquisition unit and the second computing unit, the texture maps photo that each image sheet is corresponding.

Second acquisition unit, for representing the texture dictionary of model, multiple image sheet and training in advance according to image sparse, obtains the texture sparse coefficient that each image sheet is corresponding respectively；

Second computing unit, for the texture sparse coefficient corresponding with each image sheet according to texture dictionary, calculates the texture maps photo that each image sheet is corresponding respectively.

Above-mentioned second acquisition module 303 obtains, by the second assembled unit and the second removal unit, the texture image that image to be decomposed is corresponding.

Second assembled unit, for according to each image sheet position in image to be decomposed, being combined as complete texture image by texture maps photo corresponding for each image sheet；

Second removal unit, for removing overlapping region from complete texture image, obtains the texture image that image to be decomposed is corresponding.

In embodiments of the present invention, image slices to be decomposed is become multiple image sheet；The cartoon dictionary of model, multiple image sheet and training in advance is represented according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding, represent the texture dictionary of model, multiple image sheet and training in advance according to image sparse, obtain the texture maps photo that each image sheet is corresponding；According to each cartoon image sheet, obtain the cartoon image that image to be decomposed is corresponding, and according to each texture maps photo, obtain the texture image that image to be decomposed is corresponding.Rarefaction representation is introduced in picture breakdown by the present invention, and by the different frequency composition of the cartoon dictionary of training in advance and texture dictionary picture engraving, self adaptation degree is high, striped little residue after picture breakdown.

Embodiment 4

Referring to Fig. 4, embodiments providing a kind of picture breakdown device based on rarefaction representation, this device is for performing the picture breakdown method based on rarefaction representation that above-described embodiment 2 provides.This device specifically includes:

Burst module 401, is used for treating exploded view picture and carries out burst process, obtain multiple image sheet；

Acquisition module 402, for representing, according to image sparse, the antithesis dictionary that model, multiple image sheet are corresponding with the cartoon dictionary of training in advance, obtains the cartoon image that image to be decomposed is corresponding；

Deduct module 403, for deducting cartoon image from image to be decomposed, obtain the texture image that image to be decomposed is corresponding.

Above-mentioned acquisition module 402 obtains, by the first acquiring unit, the cartoon image that image to be decomposed is corresponding with second acquisition unit.

First acquiring unit, for representing, according to image sparse, the antithesis dictionary that model, multiple image sheet are corresponding with the cartoon dictionary of training in advance, obtains the cartoon sparse coefficient that each image sheet is corresponding respectively；

Second acquisition unit, for the cartoon sparse coefficient corresponding with each image sheet according to the cartoon dictionary of training in advance, obtains the cartoon image that image to be decomposed is corresponding.

Above-mentioned second acquisition unit includes:

Computation subunit, for the cartoon sparse coefficient corresponding with each image sheet according to the cartoon dictionary of training in advance, calculates the cartoon image sheet that each image sheet is corresponding respectively；

Assembled unit, for according to each image sheet position in image to be decomposed, being combined as complete cartoon image by cartoon image sheet corresponding for each image sheet；

Removal unit, for removing overlapping region from complete cartoon image, obtains the cartoon image that image to be decomposed is corresponding.

In embodiments of the present invention, image slices to be decomposed is become multiple image sheet；Represent, according to image sparse, the antithesis dictionary that model, multiple image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon image that image to be decomposed is corresponding；From image to be decomposed, deduct cartoon image, obtain the texture image that image to be decomposed is corresponding.Rarefaction representation is introduced in picture breakdown by the present invention, and by the different frequency composition of the antithesis dictionary picture engraving of the cartoon dictionary of training in advance and correspondence thereof, self adaptation degree is high, striped little residue after picture breakdown.And introduce antithesis dictionary and can accelerate picture breakdown speed, improve picture breakdown efficiency.

The picture breakdown device based on rarefaction representation that the embodiment of the present invention provides can be the specific hardware on equipment or the software being installed on equipment or firmware etc..Those skilled in the art is it can be understood that arrive, for convenience and simplicity of description, the specific works process of system described above, device and unit, all it is referred to the corresponding process in said method embodiment.

In several embodiments provided herein, it should be understood that disclosed apparatus and method, it is possible to realize by another way.Device embodiment described above is merely schematic, such as, the division of described unit, it is only a kind of logic function to divide, actual can have other dividing mode when realizing, again such as, multiple unit or assembly can in conjunction with or be desirably integrated into another system, or some features can ignore, or do not perform.Another point, shown or discussed coupling each other or direct-coupling or communication connection can be through INDIRECT COUPLING or the communication connection of some communication interfaces, device or unit, it is possible to be electrical, machinery or other form.

The described unit illustrated as separating component can be or may not be physically separate, and the parts shown as unit can be or may not be physical location, namely may be located at a place, or can also be distributed on multiple NE.Some or all of unit therein can be selected according to the actual needs to realize the purpose of the present embodiment scheme.

It addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it is also possible to be that unit is individually physically present, it is also possible to two or more unit are integrated in a unit.

If described function is using the form realization of SFU software functional unit and as independent production marketing or use, it is possible to be stored in a computer read/write memory medium.Based on such understanding, part or the part of this technical scheme that prior art is contributed by technical scheme substantially in other words can embody with the form of software product, this computer software product is stored in a storage medium, including some instructions with so that a computer equipment (can be personal computer, server, or the network equipment etc.) perform all or part of step of method described in each embodiment of the present invention.And aforesaid storage medium includes: USB flash disk, portable hard drive, read only memory (ROM, Read-OnlyMemory), the various media that can store program code such as random access memory (RAM, RandomAccessMemory), magnetic disc or CD.

The above; being only the specific embodiment of the present invention, but protection scope of the present invention is not limited thereto, any those familiar with the art is in the technical scope that the invention discloses; change can be readily occurred in or replace, all should be encompassed within protection scope of the present invention.Therefore, protection scope of the present invention should be as the criterion with described scope of the claims.

Claims (16)

1. the picture breakdown method based on rarefaction representation, it is characterised in that described method includes:

Treat exploded view picture and carry out burst process, obtain multiple image sheet；

The cartoon dictionary of model, the plurality of image sheet and training in advance is represented according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding respectively, represent the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtain the texture maps photo that each image sheet is corresponding respectively；

According to the cartoon image sheet that described each image sheet is corresponding, obtain the cartoon image that described image to be decomposed is corresponding, and according to texture maps photo corresponding to described each image sheet, obtain the texture image that described image to be decomposed is corresponding.

2. method according to claim 1, it is characterised in that the described cartoon dictionary representing model, the plurality of image sheet and training in advance according to image sparse, obtains the cartoon image sheet that each image sheet is corresponding respectively, including:

Represent the cartoon dictionary of model, the plurality of image sheet and training in advance according to image sparse, obtain the cartoon sparse coefficient that each image sheet is corresponding respectively；

According to the cartoon sparse coefficient that described cartoon dictionary is corresponding with described each image sheet, calculate the cartoon image sheet that described each image sheet is corresponding respectively.

3. method according to claim 1, it is characterised in that the described cartoon image sheet corresponding according to described each image sheet, obtains the cartoon image that described image to be decomposed is corresponding, including:

According to described each image sheet position in described image to be decomposed, cartoon image sheet corresponding for described each image sheet is combined as complete cartoon image；

From described complete cartoon image, remove overlapping region, obtain the cartoon image that described image to be decomposed is corresponding.

4. method according to claim 1, it is characterised in that the described texture dictionary representing model, the plurality of image sheet and training in advance according to described image sparse, obtains the texture maps photo that each image sheet is corresponding respectively, including:

Represent the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtain the texture sparse coefficient that each image sheet is corresponding respectively；

According to the texture sparse coefficient that described texture dictionary is corresponding with described each image sheet, calculate the texture maps photo that described each image sheet is corresponding respectively.

5. method according to claim 1, it is characterised in that the described texture maps photo corresponding according to described each image sheet, obtains the texture image that described image to be decomposed is corresponding, including:

According to described each image sheet position in described image to be decomposed, texture maps photo corresponding for described each image sheet is combined as complete texture image；

From described complete texture image, remove overlapping region, obtain the texture image that described image to be decomposed is corresponding.

6. the picture breakdown method based on rarefaction representation, it is characterised in that described method includes:

Treat exploded view picture and carry out burst process, obtain multiple image sheet；

Represent, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon image that described image to be decomposed is corresponding；

From described image to be decomposed, deduct described cartoon image, obtain the texture image that described image to be decomposed is corresponding.

7. method according to claim 6, it is characterised in that described represent the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance according to image sparse, obtains the cartoon image that described image to be decomposed is corresponding, including:

Represent, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtain the cartoon sparse coefficient that each image sheet is corresponding respectively；

The cartoon sparse coefficient that cartoon dictionary according to training in advance is corresponding with described each image sheet, obtains the cartoon image that described image to be decomposed is corresponding.

8. method according to claim 7, it is characterised in that the described cartoon sparse coefficient corresponding with described each image sheet according to the cartoon dictionary of training in advance, obtains the cartoon image that described image to be decomposed is corresponding, including:

The cartoon sparse coefficient that cartoon dictionary according to training in advance is corresponding with described each image sheet, calculates the cartoon image sheet that described each image sheet is corresponding respectively；

According to described each image sheet position in described image to be decomposed, cartoon image sheet corresponding for described each image sheet is combined as complete cartoon image；

From described complete cartoon image, remove overlapping region, obtain the cartoon image that described image to be decomposed is corresponding.

9. the picture breakdown device based on rarefaction representation, it is characterised in that described device includes:

Burst module, is used for treating exploded view picture and carries out burst process, obtain multiple image sheet；

First acquisition module, for representing the cartoon dictionary of model, the plurality of image sheet and training in advance according to image sparse, obtain the cartoon image sheet that each image sheet is corresponding respectively, represent the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtain the texture maps photo that each image sheet is corresponding respectively；

Second acquisition module, for according to cartoon image sheet corresponding to described each image sheet, obtains the cartoon image that described image to be decomposed is corresponding, and according to texture maps photo corresponding to described each image sheet, obtains the texture image that described image to be decomposed is corresponding.

10. device according to claim 9, it is characterised in that described first acquisition module includes:

First acquiring unit, for representing the cartoon dictionary of model, the plurality of image sheet and training in advance according to image sparse, obtains the cartoon sparse coefficient that each image sheet is corresponding respectively；

First computing unit, for the cartoon sparse coefficient corresponding with described each image sheet according to described cartoon dictionary, calculates the cartoon image sheet that described each image sheet is corresponding respectively.

11. device according to claim 9, it is characterised in that described second acquisition module includes:

First assembled unit, for according to described each image sheet position in described image to be decomposed, being combined as complete cartoon image by cartoon image sheet corresponding for described each image sheet；

First removal unit, for removing overlapping region from described complete cartoon image, obtains the cartoon image that described image to be decomposed is corresponding.

12. device according to claim 9, it is characterised in that described first acquisition module includes:

Second acquisition unit, for representing the texture dictionary of model, the plurality of image sheet and training in advance according to described image sparse, obtains the texture sparse coefficient that each image sheet is corresponding respectively；

Second computing unit, for the texture sparse coefficient corresponding with described each image sheet according to described texture dictionary, calculates the texture maps photo that described each image sheet is corresponding respectively.

13. device according to claim 9, it is characterised in that described second acquisition module includes:

Second assembled unit, for according to described each image sheet position in described image to be decomposed, being combined as complete texture image by texture maps photo corresponding for described each image sheet；

Second removal unit, for removing overlapping region from described complete texture image, obtains the texture image that described image to be decomposed is corresponding.

14. the picture breakdown device based on rarefaction representation, it is characterised in that described device includes:

Burst module, is used for treating exploded view picture and carries out burst process, obtain multiple image sheet；

Acquisition module, for representing, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtains the cartoon image that described image to be decomposed is corresponding；

Deduct module, for deducting described cartoon image from described image to be decomposed, obtain the texture image that described image to be decomposed is corresponding.

15. device according to claim 14, it is characterised in that described acquisition module includes:

First acquiring unit, for representing, according to image sparse, the antithesis dictionary that model, the plurality of image sheet are corresponding with the cartoon dictionary of training in advance, obtains the cartoon sparse coefficient that each image sheet is corresponding respectively；

Second acquisition unit, for the cartoon sparse coefficient corresponding with described each image sheet according to the cartoon dictionary of training in advance, obtains the cartoon image that described image to be decomposed is corresponding.

16. device according to claim 15, it is characterised in that described second acquisition unit includes:

Computation subunit, for the cartoon sparse coefficient corresponding with described each image sheet according to the cartoon dictionary of training in advance, calculates the cartoon image sheet that described each image sheet is corresponding respectively；

Assembled unit, for according to described each image sheet position in described image to be decomposed, being combined as complete cartoon image by cartoon image sheet corresponding for described each image sheet；

Removal unit, for removing overlapping region from described complete cartoon image, obtains the cartoon image that described image to be decomposed is corresponding.