CN103745444B - Based on the non-photorealistic image rendering intent of topological tree - Google Patents

Abstract

The invention discloses a kind of non-photorealistic image rendering intent based on topological tree, comprise step: step 1, the shape topological tree of setting up original image is expressed; Step 2, according to the shape attribute of shape each in shape topological tree, keeps shape topological tree Structure, deletes the shape that shape attribute does not meet threshold requirement; Step 3, by modifying to realize image rendering and reconstruct to shape in shape topological tree.The shape topological tree that the present invention is based on image is expressed, and modifies, to reach the object that non-photorealistic image is played up, and under the prerequisite of not losing image information, greatly can improve rendering efficiency to image basic comprising unit.

Description

Based on the non-photorealistic image rendering intent of topological tree

Technical field

The invention belongs to image processing field, particularly a kind of non-photorealistic image rendering intent based on topological tree.

Background technology

Image feeling of unreality plays up (NPR, Non-PhotorealisticRendering), drawn (StylisticRendering) by translations non-photorealistic rendering or style again, refer to and utilize Practical computer teaching do not have the sense of reality as photo and have hand drawing style image technique.It originates from interactive drawing and Graphics System, creates digitized image by adding art brush effect gradually to image.Its target does not lie in the authenticity of figure, and is mainly art features, simulation artistic work (even comprising the defect in works) of performance figure or supplements as the effective of photo realism graphic.Non-photorealistic rendering does not pursue the sense of reality completely, fundamental purpose be by certain object with succinct, implicit, show with the mode of aesthetic features.

The research of playing up image feeling of unreality is hot issue always, and many researchers are devoted to automatically playing up and amendment based on style of writing ^{[1] ~ [6]}.And these work major parts all depend on basic graphical analysis, the structuring undertaken wholly or in part by image is expressed, as rim detection ^[1], linear-scale space ^[6], extracted region ^[2], conspicuousness table ^[4], laplacian pyramid ^[3]deng.

Existing image feeling of unreality is played up to be needed to be processed image by specific paintbrush model mostly, inconvenient.If can a kind of model be passed through, only revise a small amount of parameter or process, just can obtain various different feeling of unreality rendering effect, greatly will improve image processing efficiency

Reference:

[1]ThomasStrothotteandStefanSchlechtweg,Non-PhotorealisticComputerGraphics:Modeling,Rendering,andAnimation.MorganKaufmannPublishersInc.,SanFrancisco,CA,USA,2002.

[2]P.Litinowicz,Processingimagesandvideoforanimpressionisteffect.InProc.ofSIGGRAPH’97,pp.407-414,1997.

[3]G.C.P.S.B.Gooch,Artisticvision:painterlyrenderingusingcomputervisiontechniques.InProc.ofNPAR,pp.83-90,2002.

[4]S.Brooks,Mixedmediapaintingandportraiture.IEEETrans.VisualizationandComputerGraphics,Vol.12,No.5,pp.1041-1054,2007.

[5]J.P.C.a.P.M.Hall,Painterlyrenderingusingimagesalience.InProc.ofEurographicsUKConference,p.122–128,2002.

[6]A.Hertzmann,PaintByRelaxation.InProc.ofInternationalConferenceonComputerGraphics,2001.

[7]A.Hertzmann,Painterlyrenderingwithcurvedbushstrokesofmultiplesizes.InProc.ofSIGGRAPH'98,pp.453-460,1988.

Summary of the invention

For the deficiency that prior art exists, the present invention proposes a kind of can greatly improve rendering efficiency, based on the non-photorealistic image rendering intent of topological tree, the method is expressed based on the shape topological tree of image, image basic comprising unit (i.e. shape) is modified, to reach the object that non-photorealistic image is played up.

For solving the problems of the technologies described above, the present invention adopts following technical scheme:

Based on the non-photorealistic image rendering intent of topological tree, comprise step:

Step 1, the shape topological tree of setting up original image is expressed;

Step 2, according to the shape attribute of shape each in shape topological tree, keeps shape topological tree Structure, deletes the shape that shape attribute does not meet threshold requirement;

Step 3, by modifying to realize image rendering and reconstruct to shape in shape topological tree.

Step 1 comprises sub-step further:

Original color image is converted into HSV image by 1.1, gets HSV image irradiation intensity channel;

1.2 according to the intensity of illumination level of HSV image, adopts level set algorithm to obtain the basic comprising unit of image, i.e. shape;

Each shape adopts cascade structure to represent by 1.3, sets up shape topological tree and expresses, the node on behalf shape of shape topological tree, and the shape of father node representative comprises the shape of child node representative, root nodes stand entire image on image space.

Step 1.2 is specially:

According to the intensity of illumination level definition level set of HSV image, level set is level set bianry image, and it is containing one or more independently hole, adopts the pixel filling cavity that in original color image, position is corresponding with it, obtains independently shape; Continuous increase illumination intensity value, adopts level set algorithm to obtain corresponding independently shape.

According to intensity of illumination order in step 1.3, combination of shapes corresponding for each intensity of illumination is become cascade tree structure.

Adopt quick horizontal set transformation method to set up shape topological tree in step 1.3 to express.

Shape attribute described in step 2 comprises shape length breadth ratio wherein, λ ₁and λ ₂for the eigenwert of the second-order inertia matrix of shape, and λ ₁> λ ₂, delete length breadth ratio in shape topological tree and be not less than length breadth ratio threshold value ∈ ₀shape.

Shape attribute described in step 2 comprises shape area, i.e. pixel summation in shape, deletes area in shape topological tree and is not more than area threshold A ₀shape.

Shape attribute described in step 2 comprises compact shape degree wherein, A (S _i) be shape area, λ ₁and λ ₂for the eigenwert of the second-order inertia matrix of shape, and λ ₁> λ ₂, delete compactness in shape topological tree and be not more than compactness threshold value κ ₀shape.

Shape attribute described in step 2 comprises shape scale ratio wherein, A (S _i) be shape area, S _i ^rrepresent shape S _ir layer ancestors shape in shape topological tree, M is self-defining progression scope; Delete shape topological tree mesoscale ratio and be not more than scale ratio threshold alpha ₀shape.

Step 3 first in concrete scheme comprise sub-step further:

3.1 keep shape topological tree opposed configuration, adopt the Spindle Deviation θ (S of each shape respectively _i) rotated shape, wherein, (x, y) is pixel coordinate in shape; for centroid location index;

3.2 give shaped s, for shape each in shape topological tree, adjust the shape attribute of shape s respectively, make the shape attribute difference of shape in the shape attribute of shape s and shape topological tree minimum;

3.3 for shape each in shape topological tree, the mean value of all pixels color in original color image in Statistical Shape, and utilizes pixel corresponding in color average filling shape; Adopt new shape s to replace each shape in shape topological tree, obtain the shape topological tree upgraded;

3.4 based on the shape topological tree reconstructed image upgraded, and obtains shape rendering image.

The second concrete scheme of step 3 comprises sub-step further:

3.1 for shape S in shape topological tree _i, calculate its Spindle Deviation θ (S respectively _i): $\mathrm{\θ}\left(S_i\right)={\tan }^{-1}\frac{{\∫}_{S_i}(y-\stackrel{\‾}{y})\mathrm{dx}}{{\∫}_{S_i}(x-\stackrel{\‾}{x})\mathrm{dy}},$ And produce a series of satisfied ${\left(\frac{\mathrm{\Δx}}{{\mathrm{\λ}}_1}\right)}^2+{\left(\frac{\mathrm{\Δy}}{{\mathrm{\λ}}_2}\right)}^2\≤\mathrm{\ρ}$ Variable (Δ x, Δ y), wherein, λ ₁and λ ₂for the eigenwert of the second-order inertia matrix of shape, and λ ₁> λ ₂; ρ is self-defining oscillation amplitude;

3.2 produce shape s according to often organizing variable (Δ x, Δ y) _i'={ (x+ Δ x, y+ Δ y) }, and to shape s _i' carry out Random-Rotation and position and process and obtain the shape after processing;

3.3 by shape S in the shape after all process and shape topological tree _isuperposition, obtains shape s _i";

In 3.4 pairs of shape topological trees, all shapes adopt step 3.1 ~ 3.3 to process respectively, obtain amended shape topological tree;

3.5 based on amended shape topological tree reconstructed image, obtains oil painting rendering image.

The third concrete scheme of step 3 comprises sub-step further:

3.1 for shape S in shape topological tree _i, add up respectively with shape S _iin centered by each pixel, yardstick be (2t+1,2t+1) rectangular window in pixel quantity;

3.2 judge pixel quantity and rectangular window size in rectangular window, if pixel quantity is less than rectangular window area in rectangular window, then abandon the center pixel of this rectangular window;

3.3 couples of shape S _iin all pixel completing steps 3.1 ~ 3.2, shape S _ibe transformed to new shape s _i';

In 3.4 pairs of shape topological trees, each shape completing steps 3.1 ~ 3.3 respectively, obtains amended shape topological tree, based on amended shape topological tree reconstructed image, obtains fuzzy rendering image.

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

1, Shape-based interpolation topological tree expresses design of graphics as geometry, namely by cascade or the relation of inclusion expression image of shape, can recover original image based on this shape topological tree.

2, by operations such as directly modifying to shape in shape topological tree, reset, substitute, obtain amended shape topological tree, recovery is carried out to amended shape topological tree and realizes image rendering.

3, adopt new shape to replace shape in shape topological tree, obtain amended shape topological tree, recovery is carried out to amended shape topological tree and realizes image rendering.

4, expressing due to topological tree is a kind of complete expression, does not therefore lose any information of image, can adapt to any trickle operation of image; In addition, it is a kind of multi-scale expression that topological tree is expressed, and can provide the multiscale analysis of image, topological tree is a kind of simple cascade relational tree of shape, therefore image rendering operations can be converted into and operate the simple modification of shape, Fig. 1 be for original image carry out shape substitute after result.

Accompanying drawing explanation

Fig. 1 is image rendering example, and wherein, figure (a) is original color image, and figure (b) is circle rendering image, and figure (c) is line rendering image, and figure (d) is dictionary shape rendering image;

Fig. 2 is that image geometry plays up example, and wherein, figure (a) is original color image, and figure (b) is circle rendering image, and figure (c) is line rendering image, and figure (d) is dictionary shape rendering image;

Fig. 3 is that example played up by image ink, and wherein, figure (a) and figure (c) is original color image, and figure (b) and figure (d) is oil painting rendering image;

Fig. 4 is image water color rendering example, and wherein, figure (a) and figure (c) is original color image, and figure (b) and figure (d) is water color rendering image.

Embodiment

Technical scheme of the present invention is further illustrated below in conjunction with the drawings and specific embodiments.

Step 1, the shape topological tree of setting up image is expressed.

This step belongs to the known technology in this area.Concrete methods of realizing is:

For given coloured image I, be translated into HSV image I ₁, gets HSV image irradiation intensity channel, calculate topological tree Structure based on level set algorithm, and the shape topological tree of setting up image is expressed.

Level set is defined as:

χ _l(I _l)＝{p∈Ω;I _l(p)≥l}

（1）

χ ^l(I _l)＝{p∈Ω;I _l(p)≤l}

Wherein, χ _l(I _l) represent image I ₁upper level set; χ ^l(I _l) represent image I ₁sub-level set; Ω is image space; P is image slices vegetarian refreshments index; L ∈ [1 ..., L] represent the image irradiation strength level increased progressively.

Level set can regard bianry image as, namely comprises the part belonging to this level or the part not belonging to this level.Level set bianry image comprises one or more independently hole, and hole represents the pixel outside level set.To these holes, adopt the pixel that in original image I, position is corresponding with it to fill, independently shape can be obtained, the basic comprising unit of these shapes and image.By constantly increasing the value of intensity of illumination l, corresponding level set can be obtained, i.e. corresponding shape.According to the order of corresponding intensity of illumination, these combination of shapes are become tree structure, namely obtain the shape topological tree T=(S of image _i) _{i=1 ..., N}, wherein, S _ifor independently shape, i is shape indexing, and N is shape sum.

All information of image are included in independent shapes S by the shape topological tree obtained _iin, and by independent shapes S _irepresent with cascade structure.Specifically, all shapes that all level sets obtain, according to its size and relation of inclusion, form the shape topological tree of cascade, wherein, each node on behalf shape, the shape of father node representative spatially comprises the shape of child node representative, root nodes stand entire image.In addition, this shape topological tree can directly utilize quick horizontal set transformation (FastLevelSetTransformation, FLST) algorithm to obtain, and does not repeat at this.

For HSV image I ₁topological tree Structure T=(S _i) _{i=1 ..., N}, to shape S _ithe RGB triple channel color of each pixel comprised is averaging respectively, obtains shape S _icolor index c _i∈ R ³, by shape S _icolouring information be increased to topological tree Structure T=(S _i) _{i=1 ..., N}, obtain final shape topological tree T=(S _i, c _i) _{i=1 ..., N}.

Step 2, shape is selected.

For shape S _i={ (x, y); X, y ∈ Ω }, definition shape S _i(p+q) rank square μ _pq(S _i):

${\mathrm{\μ}}_{\mathrm{pq}}\left(S_i\right)=\∫\underset{S_i}{\∫}{(x-\stackrel{\‾}{x})}^p{(y-\stackrel{\‾}{y})}^q\mathrm{dxdy}---\left(2\right)$

Wherein, (x, y) is shape S _imiddle pixel coordinate; Ω is image space; P and q is respectively the norm of x and y; represent shape S _ithe location index at center, i.e. center of gravity.

$\stackrel{\‾}{x}=\frac{1}{{\mathrm{\μ}}_{00}\left(S_i\right)}\underset{S_i}{\∫\∫}\mathrm{xdxdy}---\left(3\right)$

$\stackrel{\‾}{y}=\frac{1}{{\mathrm{\μ}}_{00}\left(S_i\right)}\underset{S_i}{\∫\∫}\mathrm{ydxdy}---\left(4\right)$

In formula (3) ~ (4), μ ₀₀(S _i) be shape S _i0 rank square, i.e. shape S _iarea, it is defined as shape S _imiddle sum of all pixels.

Consider stability and the unchangeability of shape, the second moment symbolical meanings of shape is the strongest, and the most stable, and therefore the present invention utilizes the eigenvalue λ of shape second-order inertia Matrix C (s) ₁and λ ₂carry out shape selection, and λ ₁> λ ₂.

$C\left(s\right)=\left(\begin{array}{cc}{\mathrm{\μ}}_{20}\left(s\right)& {\mathrm{\μ}}_{11}\left(s\right)\\ {\mathrm{\μ}}_{11}\left(s\right)& {\mathrm{\μ}}_{02}\left(s\right)\end{array}\right)---\left(5\right)$

Namely shape is selected is shape important, representative in selected shape topological tree, and removes the assorted point in image.Shape is divided into significant and insignificant by the attribute according to shape, definition shape attribute:

Length breadth ratio area A (S _i)=μ ₀₀(S _i), compactness and scale ratio wherein, S _i ^rrepresent shape S _ir layer ancestors shape in topological tree, M is self-defining progression scope.

In above-mentioned shape attribute, length breadth ratio ∈ (S _i) define shape S _ithe ratio of major axis and minor axis, length breadth ratio shows that more greatly shape is more flat, littlely shows that shape is round; Area A (S _i) define shape S _isize; Compactness κ (S _i) define shape S _ithe intensity of middle pixel, compactness shows that more greatly shape is more irregular, and compactness is less shows that shape is close to circle; Scale ratio α (S _i) define shape S _iwith the relation of its father node shape, scale ratio shows shape S more greatly _idiffer larger with upper strata shape, scale ratio is less shows shape S _idiffer less with upper strata shape.

To shapes all in shape topological tree, Shape-based interpolation attribute judges whether each shape should retain respectively:

(1) if shape S _ilength breadth ratio ∈ (S _i) < ∈ ₀, then shape S is retained _i, namely retain length breadth ratio and be less than length breadth ratio threshold value ∈ ₀shape;

(2) if shape S _iarea A (S _i) > A ₀, then shape S is retained _i, namely Retention area is greater than area threshold A ₀shape;

(3) if shape S _icompactness κ (S _i) > κ ₀, then shape S is retained _i, namely retain compactness and be greater than compactness threshold value κ ₀shape;

(4) if shape S _iscale ratio α (S _i) > α ₀, then shape S is retained _i, namely retain scale ratio and be greater than scale ratio threshold alpha ₀shape.

In shape topological tree, whether each shape is retained to adopt one or more decision procedures in above-mentioned (1) ~ (4) to judge.The shape attribute threshold values such as above-mentioned length breadth ratio threshold value, area threshold, compactness threshold value, scale ratio threshold value rule of thumb artificially set.

Playing up and reconstruct of step 3, image.

After completing shape selection step to shape topological tree, obtain shape topological tree T', this step is by revising shape topological tree T' to reach image rendering object.The present invention proposes three kinds of image rendering schemes, comprises that image geometry redraws, image oil painting redraws and redraws with image blurring.Respectively these three kinds of image rendering schemes are described below.

(1) image geometry redraws

It is replace shape in shape topological tree T' by a shape with analog structure that image geometry redraws, and concrete methods of realizing is, to shaped s(such as: ellipse, line or dictionary shape), proceed as follows:

A () shape is biased.

In maintenance shape topological tree, between each shape, cascade connection is constant, for shape S each in shape topological tree _i, adopt following formula to calculate its Spindle Deviation θ (S respectively _i):

$\mathrm{\&theta;}\left(S_i\right)={\tan }^{-1}\frac{{\&Integral;}_{S_i}(y-\stackrel{\&OverBar;}{y})\mathrm{dx}}{{\&Integral;}_{S_i}(x-\stackrel{\&OverBar;}{x})\mathrm{dy}}---\left(6\right)$

In formula (6), (x, y) is shape S _imiddle pixel coordinate; represent shape S _ithe location index at center, can adopt formula (3) and (4) to calculate and obtain.

Utilize Spindle Deviation θ (S _i) rotated shape S _i, be specially: by shape S _iin each pixel be all multiplied by rotation transformation matrices, this is concrete implement in rotation transformation matrices $M\left(\mathrm{\&theta;}\right)=\left[\begin{array}{cc}\cos \mathrm{\&theta;}\left(S_I\right)& -\sin \mathrm{\&theta;}\left(S_I\right)\\ \sin \mathrm{\&theta;}\left(S_I\right)& \cos \mathrm{\&theta;}\left(S_I\right)\end{array}\right].$

(b) geometric transformation.

Keep the opposed configuration of shape topological tree constant, for shape S each in shape topological tree _i, calculate its area A (S respectively _i), length breadth ratio ∈ (S _i) and compactness κ (S _i).Adjustment shape s area, length breadth ratio and compactness, make shape s and shape S _ithe most similar, i.e. D (s, S in following formula _i) minimum:

D(s，S _i)＝|∈(S _i)-∈(s)|+|κ(S _ii)-κ(s)|+|A(S _i)-A(S)|(7)

Formula defines shape s and shape S in (7) _isimilarity D (s, S _i).

(c) color correction.

Keep the opposed configuration of shape topological tree constant, for shape S each in shape topological tree _i, calculate its color (S respectively _i, c _i).Concrete grammar is: to shape S _iin each pixel, add up the mean value of its RGB triple channel color in original image I respectively, utilize the color average of each pixel to shape S _imiddle respective pixel is filled; Obtained and shape S by method in step (b) _ithe most similar shape s, by each shape S in shape topological tree T' _ireplace with shape s, obtain the shape topological tree T upgraded ".

(d) Image Reconstruction.

" reconstruct the image after playing up, the recovery algorithms directly using MegaWave storehouse to provide can realize Image Reconstruction to Shape-based interpolation topological tree T, repeats no more here.Final test effect is shown in Fig. 2.

(2) image oil painting redraws

It is digital image is played up the image with painting style that image oil painting redraws.The present invention proposes shape dither method, and produce bulk redundancy information by shape shake, and produce oscillating edge movement, obtain similar fuzzy oil paint effect, concrete steps are as follows:

A () is for shape S each in shape topological tree _i, adopt formula (6) to calculate its Spindle Deviation θ (S respectively _i), and produce a series of satisfied variable (Δ x, Δ y), wherein, λ ₁and λ ₂be the eigenwert of shape second-order inertia matrix respectively, ρ is self-defining oscillation amplitude, and ρ is larger, then oscillation effect is stronger, and the image of generation is fuzzyyer.

Utilize Spindle Deviation θ (S _i) rotary variable (Δ x, Δ y), make (Δ x, Δ y) ^t=M (θ (S _i)) (Δ x, Δ y) ^t, wherein, $M\left(\mathrm{\&theta;}\left(S_i\right)\right)=\left[\begin{array}{cc}\cos \mathrm{\&theta;}\left(S_i\right)& -\sin \mathrm{\&theta;}\left(S_i\right)\\ \sin \mathrm{\&theta;}\left(S_i\right)& \cos \mathrm{\&theta;}\left(S_i\right)\end{array}\right].$

B () produces a new shape s to often organizing variable (Δ x, Δ y) _i'={ (x+ Δ x, y+ Δ y) }, (x, y) is shape S on shape topological tree _imiddle pixel; Shape s _i' carry out Random-Rotation and position processes, i.e. shape s _i' deflect an angle at random and random file position.

C () is by the mean value of all new shapes and original-shape S _isuperpose, obtain net shape s _i".

D () adopts step (a) ~ (c) to process to shapes all in shape topological tree respectively after, obtain amended shape topological tree, the recovery algorithms directly using MegaWave storehouse to provide recovers rendering image from amended shape topological tree.Value test effect is shown in Fig. 3.

(3) image blurringly to redraw

Image blurring redrawing is that logarithmic code image carries out Fuzzy Processing, produces the image of similar water color effect.The present invention proposes the image blurring of Shape-based interpolation medium filtering and redraws method, and step is as follows:

A () given medium filtering yardstick (2t+1), for shape S each in shape topological tree _i={ (x, y) }, for shape S _iall pixel (the x comprised _k, y _k), add up respectively centered by each pixel, yardstick be (2t+1,2t+1) rectangular window in pixel quantity.

If b pixel quantity is less than the half of rectangular window area in () rectangular window, then give up rectangular window center pixel.

C () is to shape S _iin after all pixel completing steps (a) ~ (b), shape S _ibe transformed to new shape s _i'.

D (), to shape each in shape topological tree completing steps (a) ~ (c) respectively, obtain amended shape topological tree, the recovery algorithms directly using MegaWave storehouse to provide recovers rendering image from amended shape topological tree.Value test effect is shown in Fig. 4.

Claims (9)

1., based on the non-photorealistic image rendering intent of topological tree, it is characterized in that, comprise step:

Step 1, the shape topological tree of setting up original image is expressed;

Step 2, according to the shape attribute of shape each in shape topological tree, keeps shape topological tree Structure, deletes the shape that shape attribute does not meet threshold requirement;

Step 3, by modifying to realize image rendering and reconstruct to shape in shape topological tree;

Step 3 comprises sub-step further:

3.1 keep shape topological tree opposed configuration, adopt the Spindle Deviation θ (S of each shape respectively _i) rotated shape, wherein, (x, y) is pixel coordinate in shape; for centroid location index;

3.2 give shaped s, for shape each in shape topological tree, adjust the shape attribute of shape s respectively, make the shape attribute difference of shape in the shape attribute of shape s and shape topological tree minimum;

3.3 for shape each in shape topological tree, the mean value of all pixels color in original color image in Statistical Shape, and utilizes pixel corresponding in color average filling shape; Adopt new shape s to replace each shape in shape topological tree, obtain the shape topological tree upgraded;

3.4 based on the shape topological tree reconstructed image upgraded, and obtains shape rendering image.

2., based on the non-photorealistic image rendering intent of topological tree, it is characterized in that, comprise step:

Step 1, the shape topological tree of setting up original image is expressed;

Step 2, according to the shape attribute of shape each in shape topological tree, keeps shape topological tree Structure, deletes the shape that shape attribute does not meet threshold requirement;

Step 3, by modifying to realize image rendering and reconstruct to shape in shape topological tree;

Step 3 comprises sub-step further:

3.1 for shape S in shape topological tree _i, calculate its Spindle Deviation θ (S respectively _i): $\mathrm{\&theta;}\left(S_i\right)={\tan }^{-1}\frac{{\&Integral;}_{S_i}(y-\stackrel{\&OverBar;}{y})dx}{{\&Integral;}_{S_i}(x-\stackrel{\&OverBar;}{x})dy},$ And produce a series of satisfied ${\left(\frac{\mathrm{\&Delta;}x}{{\mathrm{\&lambda;}}_1}\right)}^2+{\left(\frac{\mathrm{\&Delta;}y}{{\mathrm{\&lambda;}}_2}\right)}^2\&le;\mathrm{\&rho;}$ Variable (Δ x, Δ y), wherein, λ ₁and λ ₂for the eigenwert of the second-order inertia matrix of shape, and λ ₁> λ ₂; ρ is self-defining oscillation amplitude;

3.2 produce shape s according to often organizing variable (Δ x, Δ y) _i'={ (x+ Δ x, y+ Δ y) }, and to shape s _i' carry out Random-Rotation and position and process and obtain the shape after processing;

3.3 by shape S in the shape after all process and shape topological tree _isuperposition, obtains shape s _i";

In 3.4 pairs of shape topological trees, all shapes adopt step 3.1 ~ 3.3 to process respectively, obtain amended shape topological tree;

3.5 based on amended shape topological tree reconstructed image, obtains oil painting rendering image.

3., based on the non-photorealistic image rendering intent of topological tree, it is characterized in that, comprise step:

Step 1, the shape topological tree of setting up original image is expressed;

Step 2, according to the shape attribute of shape each in shape topological tree, keeps shape topological tree Structure, deletes the shape that shape attribute does not meet threshold requirement;

Step 3, by modifying to realize image rendering and reconstruct to shape in shape topological tree;

Step 3 comprises sub-step further:

3.1 for shape S in shape topological tree _i, add up respectively with shape S _iin centered by each pixel, yardstick be (2t+1,2t+1) rectangular window in pixel quantity;

3.2 judge pixel quantity and rectangular window size in rectangular window, if pixel quantity is less than rectangular window area in rectangular window, then abandon the center pixel of this rectangular window;

3.3 couples of shape S _iin all pixel completing steps 3.1 ~ 3.2, shape S _ibe transformed to new shape s' _i;

In 3.4 pairs of shape topological trees, each shape completing steps 3.1 ~ 3.3 respectively, obtains amended shape topological tree, based on amended shape topological tree reconstructed image, obtains fuzzy rendering image.

4. the non-photorealistic image rendering intent based on topological tree according to any one of claims 1 to 3, is characterized in that:

Step 1 comprises sub-step further:

Original color image is converted into HSV image by 1.1, gets HSV image irradiation intensity channel;

1.2 according to the intensity of illumination level of HSV image, adopts level set algorithm to obtain the basic comprising unit of image, i.e. shape;

Each shape adopts cascade structure to represent by 1.3, sets up shape topological tree and expresses, the node on behalf shape of shape topological tree, and the shape of father node representative comprises the shape of child node representative, root nodes stand entire image on image space.

5., as claimed in claim 4 based on the non-photorealistic image rendering intent of topological tree, it is characterized in that:

Step 1.2 is specially:

According to the intensity of illumination level definition level set of HSV image, level set is level set bianry image, and it is containing one or more independently hole, adopts the pixel filling cavity that in original color image, position is corresponding with it, obtains independently shape; Continuous increase illumination intensity value, adopts level set algorithm to obtain corresponding independently shape.

6. the non-photorealistic image rendering intent based on topological tree according to any one of claims 1 to 3, is characterized in that:

Shape attribute described in step 2 comprises shape length breadth ratio wherein, λ ₁and λ ₂for the eigenwert of the second-order inertia matrix of shape, and λ ₁> λ ₂, delete length breadth ratio in shape topological tree and be not less than length breadth ratio threshold value ∈ ₀shape.

7. the non-photorealistic image rendering intent based on topological tree according to any one of claims 1 to 3, is characterized in that:

Shape attribute described in step 2 comprises shape area, i.e. pixel summation in shape, deletes area in shape topological tree and is not more than area threshold shape.

8. the non-photorealistic image rendering intent based on topological tree according to any one of claims 1 to 3, is characterized in that:

Shape attribute described in step 2 comprises compact shape degree wherein, for shape area, λ ₁and λ ₂for the eigenwert of the second-order inertia matrix of shape, and λ ₁> λ ₂, delete compactness in shape topological tree and be not more than compactness threshold value κ ₀shape.

9. the non-photorealistic image rendering intent based on topological tree according to any one of claims 1 to 3, is characterized in that:

Shape attribute described in step 2 comprises shape scale ratio wherein, for shape area, S _i ^rrepresent shape S _ir layer ancestors shape in shape topological tree, M is self-defining progression scope; Delete shape topological tree mesoscale ratio and be not more than scale ratio threshold alpha ₀shape.