CN105392015A - Cartoon image compression method based on explicit hybrid harmonic diffusion - Google Patents

Abstract

The invention, which relates to the image processing and application field, provides a cartoon image compression method based on explicit hybrid harmonic diffusion. The method comprises the steps of feature line extraction, feature line position coding, image color coding, feature line position decoding, and image color decoding. In order to solve problems of poor practicability and the like due to limited color expression capability and long decoding time of the existing partial-differential-equation-based second-generation image compression algorithm, the invention provides a cartoon image compression method based on explicit hybrid harmonic diffusion. With simultaneous utilization of harmonic and biharmonic diffusion, diversified color changes in the image can be coded well, thereby avoiding solution of a large linear system during decoding and realizing real-time decoding. Explicit approximation is carried out on the biharmonic process by using a green function, thereby improving the decoding speed substantially. Moreover, the algorithm is clear; the effect is obvious; the real-time performance is high; and the result is robust. After industrialization, the business value of the unit flow in the mobile phone animation industry can be improved substantially and the user experience is also improved.

Description

A kind of cartoon image compression method based on explicit mixing mediation diffusion

Technical field

The present invention relates to image procossing application, particularly relate to a kind of cartoon image compression method based on explicit mixing mediation diffusion.

Background technology

Along with the development of mobile Internet, the transmission of image is more and more frequent.Because bitmap is the dot chart stored in units of pixel, capacity is comparatively large, and by the restriction of the network bandwidth, the transmission of image makes network burden large, and the consuming time of loading has had a strong impact on mutual real-time.How effectively the mobile interchange environment focused on unit discharge business and be worth, compressed image and don't affect picture quality, seem particularly important.

Traditionally, most method for compressing image (first generation compression method) all depends on the correlation technique based on classical information theory, adopt the framework of entropy code after first conversion, utilize the statistical redundancy existed in image to reach compression to view data, typically represent as JPEG2000 ^[1].But through the development of more than 20 years, such framework was tending towards ripe, and compression performance is difficult to further lifting.

In the picture, except statistical redundancy, also there is visual redundancy, namely the perception sensitivity of vision system to the loss of image zones of different of people is distinguishing.In recent years, researcher more pays close attention to and how to utilize the visual redundancy existed in image to improve compression ratio further, proposes a series of compression method, is referred to as second generation method for compressing image ^[2].These methods have merged human visual system in the design process of compression mechanism, attempt, by eliminating the mode of visual redundancy to improve code efficiency further, to keep again good visual quality simultaneously.The essence of Image Lossy Compression can be summed up as the interpolation problem of discrete data, and first generation method mostly adopts the method for such as discrete cosine transform or wavelet transform and so on.Although be also a kind of interpolation method of classics, to improve visual quality of images processing stage that partial differential equation being more used to front/rear traditionally ^[3,4].In recent years, some researchers start the core tool attempting it being used as a kind of encoding and decoding, propose the compression method respectively based on image repair (inpainting) or thermal diffusion (diffusion) ^[5,6].In addition, they notice that the vision system of people proposes the compression method of various edge maintenance to the specific sensibility at profile, edge and directional perception characteristic ^[7,8], the visual quality brought at utmost to reduce compression declines.It is pointed out that several methods based on partial differential equation proposed mostly adopt partial differential equation of second order, carrying out discrete attributes by solving Poisson's equation.But these methods can catch color areas case jumpy preferably, then effectiveness comparison is poor to change smoother region for some colors.In contrast, fourth order PDEs (biharmonic equation) better can simulate the situation of this smoothing transformation, thus also receives the concern of researcher, but the expression being vector graphics is more concentrated in current research ^[9,10].Finally, these methods are when decoding, and need to solve a huge linear system, whole process is comparatively slow, is difficult to the requirement meeting real-time decoding.

List of references:

[1]DavidTaubmanandMichaelMarcellin,JPEG2000:ImageCompressionFundamentals,StandardsandPractice.Boston:Springer,2001.

[2]M.M.Reid,R.J.Millar,andN.D.Black,“Second-generationimagecoding:anoverview,”ACMComput.Surv.,vol.29,no.1,pp.3–29,1997.

[3]IvanKopilovicandT.Szirányi，,“ArtifactReductionwithDiffusionPreprocessingforImageCompression,”Opt.Eng.,vol.44,no.2,pp.1–14,2005.

[4]F.Alter,S.Durand,andJ.Froment,“Adaptedtotalvariationforar-tifactfreedecompressionofJPEGimages,”J.Math.ImagingVis.,vol.23,no.2,pp.199–211,2005.

[5]ZhiweiXiong,XiaoyanSun,andFengWu,“Block-BasedImageCompressionWithParameter-AssistantInpainting,”IEEETrans.ImageProcess.,vol.19,no.6,pp.1651–1657,2010.

[6]IrenaGalic,JoachimWeickert,MartinWelk,AndresBruhn,AlexanderBelyaev,andHans-PeterSeidel,“ImageCompressionwithAnisotropicDiffusion,”J.Math.ImagingVis.,vol.31,pp.255–269,2008.

[7]DongLiu,XiaoyanSun,FengWu,ShipengLi,andYaqinZhang,“Imagecompressionwithedge-basedinpainting,”IEEETrans.Circuits,Syst.VideoTechnol.,vol.17,no.10,pp.1273–1287,2007.

[8]MarkusMainberger,AndrésBruhna,JoachimWeickerta,and Forchhammer,“Edge-basedcompressionofcartoon-likeimageswithhomogeneousdiffusion,”PatternRecognition.,vol.44,no.9,pp.1859–1873,2011.

[9]MarkFinch,JohnSnyder,andHuguesHoppe,“Freeformvectorgraphicswithcontrolledthin-platesplines,”ACMTrans.Graph.,vol.30,no.6,p.ArticleNo.166,2011.

[10]PeterIlbery,LukeKendall,CyrilConcolato,andMichaelMcCosker,“Biharmonicdiffusioncurveimagesfromboundaryelements,”ACMTrans.Graph.,vol.32,no.6,p.ArticleNo.219,2013.

Summary of the invention

The color representation that the object of the invention is to exist for the existing second generation image compression algorithm based on partial differential equation is limited in one's ability, the problems such as decode time is tediously long and the problem such as the practicality caused is lower, there is provided that algorithm is clear and definite, Be very effective, real-time, result robust, the business that greatly can improve mobile phone cartoon industry unit discharge after industrialization is worth, improve Consumer's Experience, promote a kind of cartoon image compression method based on explicit mixing mediation diffusion that industry is flourish.

The present invention includes following steps:

1) characteristic line extracts: calculate its laplacian image and two laplacian image to input picture, then applies non-maxima suppression method recognition feature pixel to their respectively and they are connected into characteristic line;

2) characteristic line is sorted out: introduce Laplce's counting to every bar characteristic line that laplacian image extracts, characteristic line is sampled, at each sampled point along curve method to stepping to local Laplce's minimum point to both sides, if distance is less than given threshold values, count increase by 1 to the Laplce of this curve, when Laplce's count value is greater than a half of sampling number, then judge that these lines are as Laplce's lines, all the other lines are two Laplce's lines;

3) characteristic line is position encoded: the image that all characteristic line pixels are formed is considered as a bianry image, and adopts JBIG standard to encode;

4) color of image coding: the color of image remainder is expressed as the result being in harmonious proportion and spreading, and approach with Green's function, obtain approaching the optimum Green of mediation diffusion eventually through solving a linear system, and Green's parameter that will solve the every bar characteristic line obtained stores;

5) characteristic line position decoding: the line information of JBIG coding is decoded, is loaded into after reverting to bianry image on decoded picture;

6) color of image decoding: the Green that the mediation obtained in cataloged procedure is spread is approached carry out the color value that integration obtains each pixel on image area.

In step 1) in, it is extract characteristic line in Laplace domain and two Laplace domain that described characteristic line extracts simultaneously, calculates laplacian image Δ u and two laplacian image Δ to input picture u ²u, then applies from them the identification that non-maxima suppression algorithm carries out character pixel respectively, finally character pixel is coupled together constitutive characteristic lines, then compares with Steger algorithm with conventional Cannny algorithm.

In step 2) in, described characteristic line classification can adopt a kind of method of ballot to determine that characteristic line is laplacian curve or two laplacian curve, for each point on every bar indicatrix that laplacian image takes out, normal orientation n stepping on the left of curve is until the minimum point of Laplce's absolute value; If this step distance is less than given classification threshold values, the Laplce of curve counting is added 1, otherwise two Laplce's counting is added 1, on the right side of curve, repeat this process; When the ratio of finally and if only if Laplce counting and two Laplce's counting is greater than 0.5, this curve is classified as laplacian curve; Reject the curve overlapped with the laplacian curve identified in the curve of two laplacian image extraction, remaining curve is classified as two laplacian curve; All the other lines described comprise the characteristic line on two laplacian image.

In step 4) in, described color of image coding, can carry out the distribution of color of the non-characteristic area of coded image by two Laplce's process.

In step 5) in, described characteristic line position decoding when being reconstructed image, to step 4) harmonic function u (x) that approaches of the Green that obtains is at the square region R={x ∈ (x of image ₀, x ₁), y ∈ { y ₀, y ₁on carry out integration, this integration can identify further and be expressed as the integration of Green's function core on R:

Wherein $G^L(x,y)=\frac{1}{2\mathrm{\π}}ln\left(\frac{1}{r}\right)$ For Laplce's Green's function, $G_n^L(x,x^{\′})=\frac{\∂G^L(x,x^{\′})}{\∂n\left(x^{\′}\right)},$ $G^B(x,y)=\frac{1}{8\mathrm{\π}}{r}^2\[ln\left(\frac{1}{r}\right)+1\]$ For two Laplce's Green's function, $G_n^B(x,x^{\′})=\frac{\∂G^B(x,x^{\′})}{\∂n\left(x^{\′}\right)}.$

The present invention is directed to the color representation that the existing second generation image compression algorithm based on partial differential equation exists limited in one's ability, the problems such as decode time is tediously long and the problem such as the practicality caused is lower, propose a kind of cartoon image compression method based on explicit mixing mediation diffusion, changed by the color of enriching adopting mediation and biharmonic diffusion better to exist in coded image simultaneously, avoid and solve huge linear system when decoding, thus realize real-time decoding.By adopting Green's function to carry out explicit approaching to biharmonic process, significantly improve decoding speed.Algorithm of the present invention is clear and definite, Be very effective, real-time, result robust, and the business that the method can improve mobile phone cartoon industry unit discharge greatly after industrialization is worth, and improves Consumer's Experience, promotes the flourish of industry.

Accompanying drawing explanation

Fig. 1 is overall plan flow chart of the present invention;

Fig. 2 is characteristic line leaching process figure in Fig. 1;

Fig. 3 is compressed file format figure.

Embodiment

A kind of cartoon image compression method being in harmonious proportion diffusion based on explicit mixing: the first laplacian image of calculating input image and two laplacian image extract characteristic line on their bases; Then the characteristic line obtained is extracted to laplacian image and carry out uniform sampling and Laplce's counting, determine whether these lines are laplacian curve according to count value and given threshold values, what this process is not finally classified as laplacian curve is completely considered as two laplacian curve; The image that all characteristic line pixels are formed is considered as a bianry image, and adopts JBIG standard to encode; By mediation diffusion process table coded image remainder color, approach this harmonic process with Green's function, store Green's parameter of every bar characteristic line; The line information of JBIG coding is decoded, is loaded into after reverting to bianry image on decoded picture; Finally, the Green that the mediation obtained in cataloged procedure is spread is approached carry out the color value that integration obtains each pixel on image area.

Key of the invention process has 4 points: characteristic line extracts, characteristic line is sorted out, color of image coding, color of image are decoded.Lower mask body introduction key realize details:

1, characteristic line extracts

Laplacian image Δ u and two laplacian image Δ are calculated to input picture u ²u, then applies from them the identification that non-maxima suppression algorithm carries out character pixel respectively, finally character pixel is coupled together constitutive characteristic lines.Whole leaching process can be divided into three steps:

First to the discrete two-dimentional second order local derviation Gaussian kernel of image applications one with the approximate second local derviation v of pixel each in computed image _x,x, v _y,yand v _x,y, calculate the gloomy matrix in the sea be made up of these local derviations

$H\[x,y\]=\left(\begin{array}{cc}{v}_{x,x}& v_{x,y}\\ v_{x,y}& v_{y,y}\end{array}\right)---\left(1\right)$

Characteristic value and characteristic vector just can obtain the preestimating method of an indicatrix to field;

Then the local maximum of image can then be obtained to this field of direction application non-maximum restraining core, i.e. the character pixel of constitutive characteristic curve;

Finally, from arbitrary character pixel p, expand outward from its one dimension neighborhood and find until find and its normal direction vertical direction n _⊥[p] the most close to and angle is less than the pixel q of specific threshold ^*, namely meet:

N _⊥[p] n _⊥[q ^*] > n _⊥[p] n _⊥[q ^*], and n _⊥[p] n _⊥[q ^*] > ε _c(2)

And by p and q ^*couple together.

2, characteristic line is sorted out

Adopt a kind of method of ballot to determine that characteristic line is laplacian curve or two laplacian curve. for each point on every bar indicatrix that laplacian image takes out, normal orientation n stepping on the left of curve is until the minimum point of Laplce's absolute value.If this step distance is less than given classification threshold values, the Laplce of curve counting is added 1, otherwise two Laplce's counting is added 1.This process is repeated on the right side of curve.When the ratio of finally and if only if Laplce counting and two Laplce's counting is greater than 0.5, this curve is classified as laplacian curve.Reject the curve overlapped with the laplacian curve identified in the curve of two laplacian image extraction, remaining curve is classified as two laplacian curve.

3, color of image coding

The color of other parts of entire image is expressed as the result of the mixing diffusion of the two category feature curved boundary conditions extracted with abovementioned steps.Consider that the process that biharmonic is spread contains mediation diffusion process, what therefore this process can be unified is expressed as:

Δ ²u＝0(3)

This mixing diffusion process finally can explicitly with following Green's function be approached again:

Wherein, u (x) presentation video at the color value of x, for color is at the normal direction local derviation of this point, v (x') for image is at the color Laplacian values of x, for Laplacian values is at the normal direction local derviation of this point, G ^l(x, y) is Laplce's Green's function:

$G^L(x,y)=\frac{1}{2\mathrm{\π}}ln\left(\frac{1}{r}\right),---\left(5\right)$

G ^b(x, y) is two Laplce's Green's functions:

$G^B(x,y)=\frac{1}{8\mathrm{\π}}{r}^2\[ln\left(\frac{1}{r}\right)+1\].---\left(6\right)$

In order to accurate reconstituting initial image, the weights attribute w (x') obtained in equation (1) must be solved and (comprise the normal direction local derviation of color value u (x'), color value the normal direction local derviation of color Laplacian values v (x'), color Laplacian values make with this that to approach each pixel color value that equation calculates just in time identical with the respective pixel color value of original image.By this equation discretization, to every bar characteristic line C _k(1≤k≤K, K is characteristic line quantity) carries out uniform sampling, each sampled point x' _kmweights attribute w (x' _km) pass through this curve two end points weights attribute w (x' _ks) and w (x' _ke) carry out linear interpolation and obtain w (x' _km)=(1-t _km) w (x' _ks)+t _kmw (x' _ke), t _km=l _km/ l _k(l _kmfor sampled point x' _kmto curve starting point x' _ksarc length distance, l _kfor curve C _karc length).Finally, every bar indicatrix head and the tail two end points weights attributes can be obtained by solving following linear system:

Aw＝u。(7)

Here A is M × 8K matrix (M is the pixel quantity of image), $A_{i,j+1}={\mathrm{\Δl}}_k\underset{m=1}{\overset{n_k}{\Σ}}{G}^L(x_1,{x^{\′}}_{kt})(1-t_{km}),A_{i,j+2}=-{\mathrm{\Δl}}_k\underset{m=1}{\overset{n_k}{\Σ}}\frac{\∂G^B(x_1,{x^{\′}}_{km})}{\∂n\left({x^{\′}}_{km}\right)}(1-t_{km}),$ $A_{i,j+3}={\mathrm{\Δl}}_k\underset{m=1}{\overset{n_k}{\Σ}}{G}^B(x_1,{x^{\′}}_{kt})(1-t_{km}),A_{i,j+4}=-{\mathrm{\Δl}}_k\underset{m=1}{\overset{n_k}{\Σ}}\frac{\∂G^L(x_1,{x^{\′}}_{km})}{\∂n\left({x^{\′}}_{km}\right)}{t}_{km},$ $A_{i,j+5}={\mathrm{\Δl}}_k\underset{m=1}{\overset{n_k}{\Σ}}{G}^L(x_1,{x^{\′}}_{kt})t_{km},A_{i,j+6}=-{\mathrm{\Δl}}_k\underset{m=1}{\overset{n_k}{\Σ}}\frac{\∂G^B(x_1,{x^{\′}}_{km})}{\∂n\left({x^{\′}}_{km}\right)}{t}_{km},A_{i,j+7}={\mathrm{\Δl}}_k\underset{m=1}{\overset{n_k}{\Σ}}{G}^B(x_1,{x^{\′}}_{kt})t_{km},$ Δ l _kfor the arc length distance between sampled point; W is 8K × 3 matrix, the corresponding Color Channel of each row, w _{8k, j}=u (x' _ks), $w_{8k+1,j}=\frac{\∂u\left({x^{\′}}_{ks}\right)}{\∂n\left({x^{\′}}_{ks}\right)},$ w _8k+2,j＝v(x' _ks)， $w_{8k+3,j}=\frac{\∂v\left({x^{\′}}_{ks}\right)}{\∂n\left({x^{\′}}_{ks}\right)},$ w _8k+4,j＝u(x' _ke)， $w_{8k+5,j}=\frac{\∂u\left({x^{\′}}_{ke}\right)}{\∂n\left({x^{\′}}_{ke}\right)},$ W _{8k+6, j}=v (x' _ke), u is that M × 3 comprise all image slices vegetarian refreshments.

Store as long as final, the color value of original image at each pixel can be gone out at decode phase according to these parameter reconstructs.

4, color of image decoding

In order to decode to color of image, need the square region R={x ∈ (x of harmonic function u (x) at image ₀, x ₁), y ∈ { y ₀, y ₁on carry out integration, that is:

$\mathrm{\φ}\left(R\right)=\underset{R}{\∫\∫}u\left(x\right)dx.---\left(8\right)$

This integration further can be expressed as the integration of Green's function core on R:

By mathematical derivation, can obtain analytic representation:

$F_{G^L}(R,x^{\′})=\underset{i,j\∈\{0,1\}}{\Σ}\frac{{(-1)}^{i+j}}{4\mathrm{\π}}{H}_{G^L}(\hat{x},\hat{y}),---\left(10\right)$

$H_{G^L}(\hat{x},\hat{y})=-3\hat{x}\hat{y}+\hat{x}\hat{y}\ln ({\hat{x}}^2+{\hat{y}}^2)+{\hat{x}}^{2}arctg\frac{\hat{y}}{\hat{x}}+{\hat{y}}^{2}arctg\frac{\hat{x}}{\hat{y}};---\left(11\right)$

$F_{G_n^L}(R,x^{\′})=\∫{\∫}_R{G}_n^L(x,x^{\′})dx$ Analytic representation:

$F_{G_n^L}(R,x^{\′})=\underset{i,j\∈\{0,1\}}{\Σ}\frac{{(-1)}^{i+j}}{4\mathrm{\π}}{H}_{G_n^L}(\hat{x},\hat{y},n_x,n_y),---\left(12\right)$

$H_{G_n^L}(\hat{x},\hat{y},n_x,n_y)=n_y\hat{y}arctg\left(\frac{\hat{x}}{\hat{y}}\right)+n_x\hat{x}arctg\left(\frac{\hat{y}}{\hat{x}}\right)+\frac{1}{2}(n_y\hat{x}+n_x\hat{y})ln({\hat{x}}^2+{\hat{y}}^2);---\left(13\right)$

$F_{G^B}(R,x^{\′})=\∫{\∫}_R{G}^B(x,x^{\′})dx$ Analytic representation:

$F_{G^B}(R,x^{\′})=\underset{i,j\∈\{0,1\}}{\Σ}{(-1)}^{i+j}{H}_{G^B}(\hat{x},\hat{y}),---\left(14\right)$

$H_{G^B}(\hat{x},\hat{y})=\frac{1}{48\mathrm{\π}}({\hat{x}}^4-{\hat{y}}^4)a\tan \left(\frac{\hat{x}}{\hat{y}}\right)+\frac{1}{144\mathrm{\π}}\hat{x}\hat{y}({\hat{x}}^2+{\hat{y}}^2)(11-3ln({\hat{x}}^2+{\hat{y}}^2\left)\right)---\left(15\right)$

$F_{G_n^B}(R,x^{\′})=\∫{\∫}_R{G}_n^B(x,x^{\′})dx$ Analytic representation:

$F_{G_n^B}(R,x^{\′})=\underset{i,j\∈\{0,1\}}{\Σ}{(-1)}^{i+j}{H}_{G_n^B}(\hat{x},\hat{y},n_x,n_y)---\left(16\right)$

$\begin{array}{c}{H}_{G_n^B}(\hat{x},\hat{y},n_x,n_y)=\frac{1}{48\mathrm{\π}}\left(10\hat{x}\hat{y}\right(\hat{x}{n}_x+\hat{y}{n}_y)-4{\hat{y}}^3{n}_{y}a\tan (\frac{\hat{x}}{\hat{y}})-4{\hat{x}}^3{n}_{x}a\tan (\frac{\hat{y}}{\hat{x}}\left)\right)\\ -\frac{1}{48\mathrm{\π}}({\hat{x}}^3{n}_y+3{\hat{x}}^2\hat{y}{n}_x+3\hat{x}{\hat{y}}^2{n}_y+{\hat{y}}^3{n}_x)\ln ({\hat{x}}^2+{\hat{y}}^2)\end{array}.---\left(17\right)$

Finally, by formula (10), (12), (14), (16) substitute in (9), just can decode and obtain the color value of each pixel on image.

The invention discloses a kind of cartoon image compression method based on explicit mixing mediation diffusion, comprise the following steps: 1) characteristic line extracts: its laplacian image and two laplacian image are calculated to input picture, more respectively non-maxima suppression method recognition feature pixel is applied to their and they are connected into characteristic line; 2) characteristic line is sorted out: introduce Laplce's counting to every bar characteristic line that laplacian image extracts, characteristic line is sampled, at each sampled point along curve method to stepping to local Laplce's minimum point to both sides, if distance is less than given threshold values, count increase by 1 to the Laplce of this curve, when Laplce's count value is greater than a half of sampling number, then judge that these lines are as Laplce's lines.All the other lines (comprising the characteristic line on two laplacian image) are two Laplce's lines; 3) characteristic line is position encoded: the image that all characteristic line pixels are formed is considered as a bianry image, and adopts JBIG standard to encode; 4) color of image coding: the color of image remainder is expressed as the result being in harmonious proportion and spreading, and approach with Green's function, obtain approaching the optimum Green of mediation diffusion eventually through solving a linear system, and Green's parameter that will solve the every bar characteristic line obtained stores; 5) characteristic line position decoding: the line information of JBIG coding is decoded, is loaded into after reverting to bianry image on decoded picture; 6) color of image decoding: the Green that the mediation obtained in cataloged procedure is spread is approached carry out the color value that integration obtains each pixel on image area.

The present invention is directed to the color representation that the existing second generation image compression algorithm based on partial differential equation exists limited in one's ability, the problems such as decode time is tediously long and the problem such as the practicality caused is lower, propose a kind of cartoon image compression method based on explicit mixing mediation diffusion, by the limitation adopting biharmonic diffusion technique to solve color representation ability, by adopting Green's function to carry out explicit approaching to biharmonic process, significantly improve decoding speed.Algorithm of the present invention is clear and definite, Be very effective, real-time, result robust, and the business that the method can improve mobile phone cartoon industry unit discharge greatly after industrialization is worth, and improves Consumer's Experience, promotes the flourish of industry.

Claims (5)

1., based on a cartoon image compression method for explicit mixing mediation diffusion, it is characterized in that comprising the following steps:

1) characteristic line extracts: calculate its laplacian image and two laplacian image to input picture, then applies non-maxima suppression method recognition feature pixel to their respectively and they are connected into characteristic line;

2) characteristic line is sorted out: introduce Laplce's counting to every bar characteristic line that laplacian image extracts, characteristic line is sampled, at each sampled point along curve method to stepping to local Laplce's minimum point to both sides, if distance is less than given threshold values, count increase by 1 to the Laplce of this curve, when Laplce's count value is greater than a half of sampling number, then judge that these lines are as Laplce's lines, all the other lines are two Laplce's lines;

3) characteristic line is position encoded: the image that all characteristic line pixels are formed is considered as a bianry image, and adopts JBIG standard to encode;

4) color of image coding: the color of image remainder is expressed as the result being in harmonious proportion and spreading, and approach with Green's function, obtain approaching the optimum Green of mediation diffusion eventually through solving a linear system, and Green's parameter that will solve the every bar characteristic line obtained stores;

5) characteristic line position decoding: the line information of JBIG coding is decoded, is loaded into after reverting to bianry image on decoded picture;

6) color of image decoding: the Green that the mediation obtained in cataloged procedure is spread is approached carry out the color value that integration obtains each pixel on image area.

2. a kind of based on the explicit cartoon image compression method mixing mediation diffusion as claimed in claim 1, it is characterized in that in step 1) in, it is extract characteristic line in Laplace domain and two Laplace domain that described characteristic line extracts simultaneously, calculates laplacian image Δ u and two laplacian image Δ to input picture u ²u, then applies from them the identification that non-maxima suppression algorithm carries out character pixel respectively, finally character pixel is coupled together constitutive characteristic lines, then compares with Steger algorithm with conventional Cannny algorithm.

3. a kind of based on the explicit cartoon image compression method mixing mediation diffusion as claimed in claim 1, it is characterized in that in step 2) in, the method that described characteristic line sorts out a kind of ballot of employing determines that characteristic line is laplacian curve or two laplacian curve, for each point on every bar indicatrix that laplacian image takes out, normal orientation n stepping on the left of curve is until the minimum point of Laplce's absolute value; If this step distance is less than given classification threshold values, the Laplce of curve counting is added 1, otherwise two Laplce's counting is added 1, on the right side of curve, repeat this process; When the ratio of finally and if only if Laplce counting and two Laplce's counting is greater than 0.5, this curve is classified as laplacian curve; Reject the curve overlapped with the laplacian curve identified in the curve of two laplacian image extraction, remaining curve is classified as two laplacian curve; All the other lines described comprise the characteristic line on two laplacian image.

4. a kind of based on the explicit cartoon image compression method mixing mediation diffusion as claimed in claim 1, it is characterized in that in step 4) in, described color of image coding is the distribution of color carrying out the non-characteristic area of coded image by pair Laplce's process.

5. a kind of based on the explicit cartoon image compression method mixing mediation diffusion as claimed in claim 1, it is characterized in that in step 5) in, described characteristic line position decoding when being reconstructed image, to step 4) harmonic function u (x) that approaches of the Green that obtains is at the square region R={x ∈ (x of image ₀, x ₁), y ∈ { y ₀, y ₁on carry out integration, this integration identifies further and is expressed as the integration of Green's function core on R:

wherein for Laplce's Green's function, $G^B(x,y)=\frac{1}{8\mathrm{\π}}{r}^2\[ln\left(\frac{1}{r}\right)+1\]$ For two Laplce's Green's function, $G_n^B(x,x^{\′})=\frac{\∂G^B(x,x^{\′})}{\∂n\left(x^{\′}\right)}.$