CN106709171A - Repeat mode discovery-based printed pattern generation method - Google Patents

Abstract

The invention discloses a repeat mode discovery-based printed pattern generation method. The method comprises the following steps of: carrying out repeat mode discovery on an input printed image set so as to discover an object-class image mode which repeatedly emerges in the image set; constructing a layout template for a repeat object, carrying out multi-granularity quadrilateral mesh generation and optimum layout solution on the profile of an input printed pattern through iterative loop, and calculating an object optimum layout which satisfies a printed pattern constraint condition; and calculating affine transformation of living example of the object and a hierarchical relationship between the living examples during the splicing according to the position and dimension of the template in the layout, so as to carry out living example drawing to realize the synthesis of the printed pattern.

Description

A kind of decalcomania generation method found based on repeat pattern

Technical field

The present invention relates to a kind of decalcomania generation method, belong to Computer-aided Design Technology field, specifically A kind of decalcomania generation method found based on repeat pattern.

Background technology

With the high speed development of textile printing and dyeing industry, purchaser and user are more and more diversified to the demand of decalcomania, The popular cycle of dyeing and printing products is shorter and shorter, and decalcomania change is increasingly faster, the importance more and more higher of printing pattern designing. At present, by digital printing technology, textile printing and dyeing industry can realize the printing and dyeing of the small lot decalcomania of user's customization, This imparts the feasibility that user voluntarily creates decalcomania while ordering amount threshold is reduced.But, user's creation What the intention of decalcomania was often obscured, if without reference to example, user is difficult clearly specific design content；Meanwhile, it is right Printing pattern designing process is generally too complicated for layman, and operation is more difficult；Also, due to current stamp Design is mainly directed towards mass production, and pattern contour is mostly rectangle to meet the periodicity of arrangement, and user is specific The cloth shape of demand is mostly irregular under occasion, by original decalcomania carry out cutting will certainly to pattern aesthetic property and Integrality is damaged.Therefore, extract structural heavy from existing decalcomania example using Computer-aided Design Technology Multiple object, and the decalcomania of the personalized shape for directly meeting user's use demand is generated, decalcomania can be reduced and set The difficulty of meter, enriches the application category of digital printing technology.

There are many researchers to attempt carrying out from different angles the generation of area of computer aided decalcomania, example at present Such as：Document 1：Zhao Haiying, Peng Hong, Yang Yifan, the positive light of Xu are based on carpet pattern generation method [J] computer aided manufacturings of topology configuration Help design and graphics journal, 2013,25 (4):The Xinjiang ethnic groups carpet style decalcomania synthesis side proposed in 502-509. Method；Document 2：Djibril M O,Thami R O H.IsLamic geometrical patterns indexing and classification using discrete symmetry groups[J].Journal on Computing and Cultural Heritage,2008,1(2):The Islamic theme novel pattern generation method proposed in Article No.10.； Document 3：Research [J] the computer research of the creation of Bu Jiajun, Xu Duanqing, Chen Chun, Ma Lingzhou moire patterns and preparation method and hair Exhibition, 2001,38 (1):Moire pattern generation method proposed in 105-110. etc..These methods are all from decalcomania style or reality The angle of example content is set out, inherent characteristicses layout layout and generating algorithm according to example, and single algorithm is to different type Decalcomania adaptability it is poor, it is impossible to meet to the reconstruction of various decalcomania examples and synthesis.

On the other hand, document 4：The composition art and its computer of the small inscription fractal patterns of king realize [J] area of computer aided Design and graphics journal, 2001,13 (1):The fractal pattern formative method proposed in 83-86.；Document 5：Wang Qi, Liu Hong, Nie Application [J] CAD and graphics journal, 2010,22 (1) of the brilliant evolution art in wall paints design: 24-29；Cycloid configuration generating algorithm based on the art of evolving of middle proposition etc..These methods are all from the angle of geometric space layout Set out, by the primitive geometric element to representative pattern example carry out logic repeatedly, conversion or the combined crosswise of position formed Different layout results symmetrically or non-symmetrically, therefore the decalcomania outline of generation is all fixed, is mostly rectangle, it is impossible to To the more uniform decalcomania layout of user-defined non-rectangular pattern profile generation example distribution.

The content of the invention

Goal of the invention：The technical problems to be solved by the invention are directed to the deficiencies in the prior art, there is provided one kind is based on weight The automation decalcomania generation method that complex pattern finds, for supporting that area of computer aided finds to repeat mould from decalcomania example Formula, and reproduce decalcomania.

In order to solve the above-mentioned technical problem, the invention discloses a kind of decalcomania generation side found based on repeat pattern Method, comprises the following steps：

Step 1, repeating objects find：Input picture collection, is partitioned into foreground object, to the foreground object that is partitioned into iteratively Similarity mode between any two is carried out, so that the object level image model repeated in calculating image set.

Repeating objects are arranged by step 2：The template being made up of quadrilateral mesh to the repeating objects construction for excavating, And by iterative cycles, the decalcomania profile to giving carries out many granularity quadrilateral mesh subdivisions and discrete compact arrangement is asked Solution, calculates the object optimal location for meeting decalcomania constraints.

Step 3, synthesizes decalcomania：Template position and yardstick in object optimal location, calculate layout real The affine transformation of example synthesis, and hierarchical relationship when there is covering between layout example, so as to be laid out the drafting of example, To realize the synthesis of decalcomania.

Step 1 of the present invention is comprised the following steps：

Step 1-1, Object Segmentation：Input picture collection S, calculates each saliency map of image I in image set S PSaliency, and a mask image, background before mask image is carried out into image as initialization condition are generated to its thresholding Segmentation, the parameter setting of wherein thresholding is as follows：

Wherein PSaliency (p) is the value of pixel p in saliency map picture, and PMask (p) is pixel p in mask image Value, GC_FGD represents PMask (p) for foreground pixel, and GC_BGD represents PMask (p) for background pixel, and GC_PRFGD is represented PMask (p) may be foreground pixel, and the interior zone that each the outermost layer profile in preceding background segment result is included is used as one The mask OMask of individual foreground object, if OMask (p)=0, pi is background pixel, and otherwise pi is foreground pixel, foreground pixel Region is foreground object OImg in corresponding image I.All foreground objects composition foreground object set O of image I_I。

Step 1-2, similitude matching：To the foreground object set O of arbitrary image I in image set S_I, appoint and take two of which Foreground object OImg_iAnd OImg_j, 0≤i≤j≤N_I, N_IIt is O_IMiddle foreground object sum, calculates foreground object OImg respectively_iWith OImg_jSet of characteristic points F_iAnd F_j.For foreground object OImg_iSet of characteristic points F_iIn each characteristic point f_i,kf, 0≤kf ≤n_i, wherein n_iIt is F_iComprising feature point number, calculate foreground object OImg_jSet of characteristic points F_jIn closest therewith spy Levy a littlef_j,l∈F_j, and secondary neighbouring characteristic pointf_j,l∈ F_j-f_j,lIf,Then by f_i,kfAnd f_j,lThe point for regarding one group of correct matching as is right.If set of characteristic points F_iIn just Really the characteristic point of matching is to number n_i,j>0.3n_i, then foreground object OImg_iWith OImg_jIt is mutually matched, is classified as cluster, and compare Compared with its set of characteristic points F_iAnd F_jIn feature point number, selection feature point number relatively large number of foreground object generation therebetween The repeating objects of table this cluster, repeating objects OImg_RRepresent, and in foreground object set O_IMiddle deletion foreground object OImg_i And OImg_j.Iteratively to repeating objects OImg_RWith foreground object set O_IIn remaining foreground object carry out similitude successively Match somebody with somebody, and update repeating objects OImg_RWith the foreground object for deleting matching.

Step 1-3, judges whether to complete：Check foreground object set O_IWhether it is sky, if it is empty, or foreground object set O_IMiddle foreground object number is equal to 1, then go to step 2-1；Otherwise, 1-2 is gone to step.

Step 2 of the present invention is comprised the following steps：

Step 2-1, generates template：To representational repeating objects OImg_RCorresponding mask image OMask_RCarry out uniform Rectangular mesh subdivision, the granularity of subdivision isWherein Width_R、Height_RRespectively mask figure As OMask_RWide and height.The grid generated after subdivision is designated as M_R, calculate mask image OMask_RIn each dough sheet b_R,kbIt is corresponding The foreground area ratio R in region_R,kb：

Wherein, OMask_RP () is mask image OMask_RThe value of middle pixel p, if R_R,kb<25%, then by dough sheet b_R,kbFrom M_R Middle deletion.

Template T_RTopological structure be defined as follows：To grid M_RIn all nodes set up by V_BIn all nodes compositions Communication path and V_IIn all nodes composition minimum spanning tree, wherein V_BIt is grid M_RIn boundary node set, V_IIt is grid M_RIn non-boundary node set, calculate in communication path and minimum spanning tree each node motion to the coordinate side of next node To being stored as sequence L_R, L_R=(l₁,l₂,…l_n), l_k∈ { 0,1,2,3,4 }, wherein 1≤k≤n-1, n are grid M_RIn section Point sum, l_kRepresent k-th node v in communication path_kWith+1 vertex v of kth_k+1Between spatial relation, its value be 0 generation Table x_k>x_k+1, it is 1 to represent y_k>y_k+1, it is 2 to represent x_k≤x_k+1, it is 3 to represent y_k≤y_k+1, wherein x_k、y_kV is represented respectively_kHorizontal seat Mark and ordinate, x_k+1、y_k+1V is represented respectively_k+1Abscissa and ordinate, while calculating all branches section in minimum spanning tree Point is in L_RIn sequence number, be stored as sequence B_R, L_RAnd B_RCollectively constitute repeating objects OImg_RCorresponding template T_RTopological structure, l₁Corresponding grid node is anchor point；

Step 2-2, quadrilateral mesh subdivision：What user was given is carried out for generating the two-dimensional silhouette region of decalcomania Quadrilateral mesh subdivision, subdivision granularity Q is initialized as 100, generates the quadrilateral mesh M of profile.

Step 2-3, discrete compact arrangement：Calculate each template T_RAll layout realities for meeting topological structure in grid M Example T_R,a, wherein a is the position of anchor point in the communication path of the layout example in grid M, T_R,a∈ { 0,1 }, wherein T_R,a= 1 represents during the layout example appears in final layout result, T_R,a=0 represents layout example quilt in final layout result Abandon, layout example T is calculated by equation below_R,aWeight w_R,a：

Wherein rand is any random integers, and b is any dough sheet in grid M, and following about to layout example addition Beam condition：

WhereinIt is average foreground object number, N contained by every image in image set_TIt is template number.By calculating prospect Example under specifying constraint discretization global optimum layout, obtain repeating objects current grid and constraint under the conditions of Optimal discrete compact arrangement, i.e. value are 1 prospect example T_R,aSet.

Step 2-4, judges whether to receive：The layout example T that the value for obtaining is 1 is solved in statistic procedure 2-3_R,aNumber C, ifThen go to step 3-1；Otherwise, update the value of Q in step 2-2 for Q 1.2, and go to step 2-2.

Step 3 of the present invention is comprised the following steps：

Step 3-1, object transformation：Value to being produced in each step 2-3 is 1 layout example T_R,aCorresponding repetition Object is rotated and is scaled equiaffine conversion.OImg_REntering behavior carries out angle for r (OImg_R, rotation and ratio a) are s (OImg_R, a) it is the scaling of ratio, the rendered object OImg after generation affine transformation_rs,a, and determine the drafting center on painting canvas v_c,a；

Anglec of rotation r (OImg_R, computing formula a) is：

Wherein v_kv,aIt is layout example T_R,aIn grid M_RIn any summit,Respectively v_kv,aIt is horizontal, vertical Coordinate, v_c,aIt is layout example T_R,aThe centre of form,Respectively v_c,aAbscissa and ordinate, v_kvIt is template T_RIn Any summit,Respectively v_kvAbscissa and ordinate, v_c,RIt is template T_RThe centre of form,Respectively v_c,R Abscissa and ordinate, n_vRIt is template T_RThe node total number that includes of grid.Pantograph ratio s is calculated by equation below (OImg_R,a)：

Wherein e_mIt is template T_RGrid in any a line.

Step 3-2, hierarchical ranking：The rendered object area that all layout examples are crossed according to corresponding affine transformation is from small To big sequence, formation sequence Q, wherein for arbitrary placement example T_R,a, the real area S (T in painting canvas_R,a) be：

S(T_R,a)=s (OImg_R,a)×p∈OMask_R[OMask_R(p)=255]

Step 3-3, objects draw：According to the sequence Q produced in step 3-2, product in all step 3-1 is drawn on painting canvas Rendered object after raw affine transformation, wherein for arbitrarily drawn object OImg_rs,a, position is drawn with v_c,aCentered on, with reality The generation of existing decalcomania.

Beneficial effect：The present invention has advantages below：First, the present invention can concentrate discovery to meet in given graphic image The pattern sample of decalcomania demand, that is, the subject area for repeating, for generating new decalcomania.Secondly, the present invention The decalcomania of object compact arrangement can be quickly realized on the premise of the abundant pattern contour of User Defined is met.Finally, In the decalcomania of the present invention generation same target dispersed distribution in a given area, and different object distributed quantity more Uniformly.

Brief description of the drawings

The present invention is done with reference to the accompanying drawings and detailed description further is illustrated, it is of the invention above-mentioned or Otherwise advantage will become apparent.

Fig. 1 is handling process schematic diagram of the invention.

Fig. 2 a and Fig. 2 b are the input graphic image collection example schematic diagrams of embodiment.

Fig. 2 c and Fig. 2 d are the schematic diagrames that Object Segmentation is carried out to the input picture collection of embodiment.

Fig. 3 a are that ground similarity mode is iterated to the input picture collection of embodiment, and repeating objects shows obtained from It is intended to.

Fig. 3 b are the schematic diagrames of the corresponding mask image of foreground object of Fig. 3 a.

Fig. 4 a are the schematic diagrames that corresponding templates grid is generated to the foreground image of Fig. 3 a.

Fig. 4 b are the decalcomania outline drawings of embodiment.

Fig. 4 c are the schematic diagrames that quadrilateral mesh subdivision is carried out to the decalcomania profile of Fig. 4 b.

Fig. 4 d are, to Fig. 4 c quadrilateral mesh, the schematic diagram of discrete compact arrangement to be carried out in the template of Fig. 4 a.

Fig. 5 draws the schematic diagram of decalcomania on the basis of discrete compact arrangement.

Specific embodiment

As shown in figure 1, disclosed by the invention is a kind of decalcomania generation method found based on repeat pattern, specific bag Include following steps：

Step one, repeating objects find：To existing image set, foreground object is partitioned into, and two are made iteratively to it Similarity mode between two, so that the object level image model repeated in calculating image set.

Step 2, repeating objects arrangement：The template being made up of quadrilateral mesh to the repeating objects construction for excavating, and pass through Iterative cycles, the decalcomania profile to giving carries out many granularity quadrilateral mesh subdivisions and discrete compact arrangement is solved, and calculates Go out to meet the object optimal location of decalcomania constraints.

Step 3, decalcomania synthesis：Template position and yardstick in layout, calculate layout example synthesis Affine transformation, and hierarchical relationship when there is covering between layout example, so as to be laid out the drafting of example, to realize print The synthesis of floral pattern.

Lower mask body introduces the main flow of each step：

Step 1, repeating objects find：

Repeating objects discovery procedure carries out the other repeat pattern of object level to input picture collection and excavates, and in each repetition A representative object is chosen in the cluster of object, for decalcomania generation provides pattern source.Including following steps：

Step 11, Object Segmentation：Document 6 is used to each image I in image set S：Efficient Salient Region Detection with Soft Image Abstraction[C].International Conference on Computer Vision,2013:Global Uniqueness methods generation saliency map described in 1529-1536. PSaliency, and a mask image, background before mask image is carried out into image as initialization condition are generated to its thresholding Segmentation, the parameter setting of wherein thresholding is as follows：

Wherein PSaliency (p) is the value of pixel p in saliency map picture, and PMask (p) is pixel p in mask image Value, GC_FGD represents PMask (p) for foreground pixel, and GC_BGD represents PMask (p) for background pixel, and GC_PRFGD is expression PMask (p) may be foreground pixel, and preceding background segment method uses document 7：“Grabcut”–Interactiveforegroun d extraction using iterated graph cuts[J].ACMTransactions onGraphics2004,23 (3):The Grabcut algorithms proposed in 309-314., iteration carries out four segmentations, carries out one to mask image after segmentation every time Secondary expansion and etching operation.The interior zone that each outermost layer profile in preceding background segment result is included is used as a prospect The mask OMask of object, the situation of background area is included with foreground area in adapting to segmentation result, if OMask (p)=0, P is background pixel, and otherwise p is foreground pixel, and region is foreground object OImg in the corresponding image I of foreground pixel.The institute of image I Foreground object set O is constituted by foreground object_I。

Step 12, similitude matching：To the foreground object set O of arbitrary image I in image set_I, appoint and take two foreground objects OImg_iAnd OImg_j, 0≤i≤j≤N_I, N_IIt is O_IMiddle foreground object sum, calculates set of characteristic points F respectively_iAnd F_j, feature point extraction Method uses document 8：Distinctive Image Features from Scale-Invariant Keypoints[J] .International Journal of Computer Vision,2004,60(2):The SIFT feature description that 91-110. is proposed Son.For F_iIn each characteristic point f_i,kf, 0≤kf≤n_i, wherein n_iIt is F_iComprising feature point number, calculate F_jIn therewith most Neighbouring characteristic pointf_j,l∈F_j, and secondary neighbouring characteristic point f_j,l∈F_j-f_j,lIf,Then by f_i,kfAnd f_j,lThe point for regarding one group of correct matching as is right.If F_iIn correct The characteristic point matched somebody with somebody is to number n_i,j>0.3n_i, then OImg_iWith OImg_jIt is mutually matched, is classified as cluster, and compare its characteristic point Set F_iAnd F_jIn feature point number, selection therebetween the relatively large number of foreground object of feature point number represent this cluster Repeating objects, repeating objects OImg_RRepresent, and in O_IMiddle deletion OImg_iAnd OImg_j.Iteratively to repeating objects OImg_RWith Foreground object set O_IIn remaining foreground object carry out similitude matching successively, and update repeating objects OImg_RMatched with deleting Foreground object.

Step 13, judges whether to complete：Check O_IWhether it is sky, if it is empty, or O_IMiddle foreground object number is equal to 1, then turn Step 21；Otherwise, 12 are gone to step.

Step 2, repeating objects arrangement：

During repeating objects arrangement, the template of the repeating objects generation quadrilateral mesh composition first to finding, while Decalcomania profile to being input into carries out the quadrilateral mesh subdivision and discrete compact arrangement of many granularities, by a mistake for iteration Journey, realization meets the compact arrangement result of repeating objects quantitative requirement.Including following steps：

Step 21, template generation：First to representative repeating objects OImg_RCorresponding mask image OMask_RCarry out uniform Rectangular mesh subdivision, the granularity of subdivision isWherein Width_R、Height_RRespectively OMask_RWide and height, so as to by OMask_RIt is divided into a series of completely the same rectangular block of shapes.For the net generated after subdivision Lattice M_R, in order that the shape in template fitting repeating objects region, calculates each dough sheet b in mask image_R,kbBefore corresponding region Scape area ratio R_R,kb, computing formula is as follows：

Wherein OMask_RP () is the value of mask image pixel p, if R_R,kb<25%, then by dough sheet f_R,kbFrom M_RMiddle deletion.

Due to M_RThere is communication path in middle boundary node, non-boundary node may not exist communication path, therefore to grid M_R In all nodes set up by V_BIn all nodes composition communication path and V_IIn all nodes composition minimum spanning tree, its Middle V_BIt is grid M_RIn boundary node set, V_IIt is grid M_RIn non-boundary node set, calculate communication path and most your pupil Each node motion of Cheng Shuzhong is stored as sequence L to the coordinate direction of next node_R, L_R=(l₁,l₂,…l_n), l_k∈{0,1, 2,3,4 }, 1≤k≤n-1, wherein n are grid M_RIn node total number, l_kRepresent k-th node v in communication path_kWith kth+1 Individual vertex v_k+1Between spatial relation, its value represents x for 0_k>x_k+1, it is 1 to represent y_k>y_k+1, it is 2 to represent x_k≤x_k+1, it is 3 Represent y_k≤y_k+1, wherein x_k、y_kV is represented respectively_kAbscissa and ordinate, x_k+1、y_k+1V is represented respectively_k+1Abscissa and vertical Coordinate, while all branch nodes are in L in calculating minimum spanning tree_RIn sequence number, be stored as sequence B_R, L_RAnd B_RCollectively constitute Repeating objects OImg_RCorresponding template T_RTopological structure, l₁Corresponding grid node is anchor point；

Step 22, quadrilateral mesh subdivision：What user was given is carried out for generating the two-dimensional silhouette region of decalcomania Quadrilateral mesh subdivision, subdivision algorithm is using using document 9：Data-Driven Interactive Quadrangulation [J].Acm Transactions on Graphics,2015,34(4):Subdivision algorithm in 1-10., subdivision granularity Q initialization It is 100, generates the quadrilateral mesh M of profile.

Step 23, discrete compact arrangement：Calculate each template T_RAll layout examples for meeting topological structure in grid M T_R,a, wherein a is the position of anchor point in the communication path of the layout example in grid M, T_R,a∈ { 0,1 }, wherein T_R,a=1 Represent during the layout example appears in final layout result, T_R,a=0 represents the layout example is thrown in final layout result Abandon, layout example T is calculated by equation below_R,aWeight w_R,a：

Wherein rand is any random integers, and f is any dough sheet in M.Using document 10：Computing Layouts with Deformable Templates[J],Acm Transactions on Graphics,2014,33(4):In 70-79. Discretization global optimum placement algorithm, be calculated as follows the globally optimal solution of Objective function：

Wherein b_kbIt is any one dough sheet in M,It is average foreground object number, N contained by every image in image set_T It is template number, first function is the object function of discrete optimal location, and second and third inequality is constraint inequality, can Solve the prospect example T that optimal discrete compact arrangement of the repeating objects under the conditions of current grid and constraint, i.e. value are for 1_R,aCollection Close.

Step 24, judges whether to receive：The layout example T that the value for obtaining is 1 is solved in statistic procedure 23_R,aNumber C, IfThe requirement that layout result meets decalcomania layout is represented, then goes to step 31；Otherwise, quadrilateral mesh subdivision is reduced Granularity, that is, update the value of Q in step 22 for Q 1.2, and go to step 22.

Step 3, decalcomania synthesis：

In decalcomania building-up process, the result first according to discrete compact arrangement calculates each repeating objects and is plotted in cloth The affine transformation to be carried out in the corresponding canvas location of office's example, and the real area in painting canvas is plotted according to layout example The hierarchical relationship between object is calculated, so as to realize the synthesis of decalcomania.Including following steps：

Step 31, object transformation：Value to being produced in each step 23 is 1 layout example T_R,aCorresponding repetition is right As being rotated and being scaled equiaffine conversion, OImg_REntering behavior carries out angle for r (OImg_R, rotation and ratio a) are s (OImg_R, a) it is the scaling of ratio, the rendered object OImg after generation affine transformation_rs,a, and determine the drafting center on painting canvas v_c,a；The numerical value of affine transformation is solved by the concrete condition of template instances in grid M and obtained.Anglec of rotation r (OImg_R, meter a) Calculating formula is：

Wherein v_kv,aIt is layout example T_R,aIn grid M_RIn any summit,Respectively v_kv,aIt is horizontal, vertical Coordinate, v_c,aIt is layout example T_R,aThe centre of form,Respectively v_c,aAbscissa and ordinate, v_kvIt is template T_RIn Any summit,Respectively v_kvAbscissa and ordinate, v_c,RIt is template T_RThe centre of form, Respectively v_c,RAbscissa and ordinate, n_vRIt is template T_RThe node total number that includes of grid.Pantograph ratio s is calculated by equation below (OImg_R,a)：

Wherein e_mIt is template T_RGrid in any a line.

Step 32, hierarchical ranking：Situation about blocking is there may be between the corresponding repeating objects of layout example, for existing The larger repeating objects of region area are plotted in upper strata by two repeating objects of hiding relation, meet near big and far smaller perspective Relation.Therefore, all layout examples are sorted from small to large according to the rendered object area that corresponding affine transformation is crossed, generates sequence Row Q, for arbitrary placement example T_R,a, the real area S (T in painting canvas_R,a) be：

Step 33, objects draw：According to the sequence Q produced in step 32, generation in all step 3-1 is drawn on painting canvas Affine transformation after rendered object, wherein for arbitrarily drawn object OImg_rs,a, position is drawn with v_c,aCentered on, to realize The generation of decalcomania.

Embodiment

In the present embodiment, to the image set being made up of two width printing images of Fig. 2 a and Fig. 2 b inputs, Fig. 2 c and Fig. 2 d are The mask image generated after Object Segmentation, wherein each white portion represents a foreground object, it is made iteratively two-by-two Between match and similitude and find repeating objects described in Fig. 3 a.The different repeating objects of 7 classes are found that in the present embodiment altogether, respectively It is five kinds of rabbits of different shape, deer and five-pointed stars, Fig. 3 b are the corresponding mask image of repeating objects of Fig. 3 a.Fig. 4 a be based on The template mesh of the quadrilateral mesh composition of Fig. 3 b generations.Fig. 4 a are the profile of the target decalcomania of user input, and it is entered Row quadrilateral mesh subdivision generates grid described in Fig. 4 c, and based on quadrilateral mesh and template described in Fig. 4 a carries out discrete tight cloth Office to solve and be laid out optimal solution described in generation Fig. 4 d, generates 46 layout examples in the present embodiment altogether.Fig. 5 is the stamp figure of synthesis Case result.

Specific implementation process is as follows：

In step one, every image to input picture collection shown in Fig. 2 a and Fig. 2 b calculates saliency map, and according to notable Degree figure thresholding generates a mask image as initialization condition, and background segment, knot before image are carried out using Grabcut algorithms Fruit is as shown in Fig. 2 c and Fig. 2 d；To each image in Fig. 2 c and Fig. 2 d, iteratively match similar between wherein each two object Property, as a result the most object of selected characteristic point number such as schemes as repeating objects in foreground object of every cluster coupling number more than 1 Shown in 3a, the corresponding mask image of each repeating objects is as shown in Figure 3 b.

In step 2, repeating objects are arranged according to given decalcomania profile.First to the weight shown in Fig. 3 b The template of multiple object mask generation quadrilateral mesh composition, network result is as shown in fig. 4 a；Meanwhile, to defeated shown in Fig. 4 b Entering decalcomania profile carries out quadrilateral mesh subdivision, and subdivision granularity is initialized as 100, and in the quadrilateral mesh of subdivision generation Template shown in upper discrete compact arrangement Fig. 4 a.If the layout example number of each template reaches print in the result of discrete compact arrangement The requirement of floral pattern, then using the layout result as generation decalcomania layout；Subdivision granularity is otherwise reduced, again to print Floral pattern profile carries out quadrilateral mesh subdivision and calculates discrete compact arrangement.Fig. 4 c are to meet stamp based on Fig. 4 a generations Quadrilateral mesh under the subdivision granularity of pattern requirement, Fig. 4 d are based on the mould shown in the quadrilateral mesh shown in Fig. 4 c and Fig. 4 a The discrete compact arrangement result that plate is calculated is illustrated, wherein black silhouette of different shapes represents different template types, wheel Gray area in exterior feature represents the layout example centered on position of form center, by rotating and scale in discrete compact arrangement.

In step 3, to the layout example in discrete compact arrangement shown in Fig. 4 d according to the real area in painting canvas from big To small sequence, and based on each layout example positions, the anglec of rotation and scaling in Fig. 4 d, sequentially painted on painting canvas according to this Rendered object after repeating objects affine transformation shown in drawing 3a, the decalcomania of synthesis is as shown in Figure 5.

The invention provides a kind of decalcomania generation method found based on repeat pattern, the technical scheme is implemented Method and approach it is a lot, the above is only the preferred embodiment of the present invention, it is noted that general for the art For logical technical staff, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improve and Retouching also should be regarded as protection scope of the present invention.Each part being not known in the present embodiment can use prior art to be subject to reality It is existing.

Claims (7)

1. it is a kind of based on repeat pattern find decalcomania generation method, it is characterised in that comprise the following steps：

Step 1, finds repeating objects：Input picture collection, is partitioned into foreground object, and the foreground object to being partitioned into is made iteratively Similarity mode between any two, so that the object level image model repeated in calculating image set；

Repeating objects are arranged by step 2：The template being made up of quadrilateral mesh to the repeating objects construction for excavating, and lead to Iterative cycles are crossed, the decalcomania profile to giving carries out many granularity quadrilateral mesh subdivisions and discrete compact arrangement is solved, meter Calculate the object optimal location for meeting decalcomania constraints；

Step 3, synthesizes decalcomania：Template position and yardstick in object optimal location, calculate layout example and close Into affine transformation, and exist covering when layout example between hierarchical relationship, so as to be laid out the drafting of example, realize The synthesis of decalcomania.

2. method according to claim 1, it is characterised in that step 1 is comprised the following steps：

Step 1-1, Object Segmentation：Input picture collection S, calculates any one saliency map of image I in image set S PSaliency, and a mask image is generated to its thresholding, each the outermost layer profile in preceding background segment result is surrounded Interior zone as a mask OMask for foreground object, in mask image in the corresponding image I of foreground area region for preceding All foreground objects composition foreground object set O of scape object OImg, image I_I；

Step 1-2, similitude matching：To the foreground object set O of arbitrary image I in image set S_I, appoint and take two of which prospect pair As OImg_iAnd OImg_j, foreground object OImg is calculated respectively_iAnd OImg_jSet of characteristic points F_iAnd F_j, 0≤i≤j≤N_I, N_IFor preceding Scape object set O_IMiddle foreground object sum；

For foreground object OImg_iSet of characteristic points F_iIn each characteristic point f_i,kf, 0≤kf≤n_i, n_iIt is characterized point set F_iComprising feature point number, calculate foreground object OImg_jSet of characteristic points F_jIn with characteristic point f_i,kfMost Euclidean distance is most adjacent Near characteristic point f_j,l, and the neighbouring characteristic point f of Euclidean distance time_j,mIf,Then by f_i,kfAnd f_j,lRegard as The point of one group of correct matching is right, the feature point number that point set is included is characterized, if set of characteristic points F_iIn correct matching spy Levy a little to number n_i,j>0.3n_i, then foreground object OImg_iWith OImg_jIt is mutually matched, is classified as cluster, and compare its characteristic point Set F_iAnd F_jIn feature point number, selection therebetween the relatively large number of foreground object of feature point number represent this cluster Repeating objects, repeating objects OImg_RRepresent, in foreground object set O_IMiddle deletion foreground object OImg_iAnd OImg_j, iteration Ground is to repeating objects OImg_RWith foreground object set O_IIn remaining foreground object carry out similitude matching successively, and update repetition Object OImg_RWith the foreground object for deleting matching；

Step 1-3, judges whether to complete：Check foreground object set O_IWhether it is sky, if it is empty, or foreground object set O_IIn Foreground object number is equal to 1, then go to step 2-1；Otherwise, 1-2 is gone to step.

3. method according to claim 2, it is characterised in that step 2 is comprised the following steps：

Step 2-1, generates template：To representational repeating objects OImg_RCorresponding mask image OMask_RCarry out uniform rectangle Mesh generation, the granularity of subdivision isWherein Width_R、Height_RRespectively mask image OMask_RWide and height, after subdivision generate grid be designated as M_R, calculate mask image OMask_RIn each dough sheet b_R,kbCorresponding area The foreground area ratio R in domain_R,kb：

$R_{R,kb}=\frac{{\mathrm{\&Sigma;}}_{p\&Element;b_{R,kb}}\&lsqb;{\mathrm{OMask}}_R\left(p\right)=255\&rsqb;}{b_{R,kb}},$

Wherein, OMask_RP () is mask image OMask_RThe value of middle pixel p, if R_R,kb<25%, then by dough sheet b_R,kbFrom grid M_R Middle deletion, and set up repeating objects OImg_RCorresponding template T_R, T_RTopological structure in first appearance grid vertex be anchor Point；

Step 2-2, quadrilateral mesh subdivision：What user was given carries out four sides for generating the two-dimensional silhouette region of decalcomania Shape mesh generation, subdivision granularity Q is initialized as 100, generates the quadrilateral mesh M of profile；

Step 2-3, discrete compact arrangement：Calculate each template T_RAll layout example T for meeting topological structure in grid M_R,a, Wherein a is the position of anchor point in the communication path of the layout example in grid M, T_R,a∈ { 0,1 }, wherein T_R,a=1 represents The layout example is appeared in final layout result, T_R,a=0 represents the layout example is abandoned in final layout result, Layout example T is calculated by equation below_R,aWeight w_R,a：

$w_{R,a}=(1+\frac{rand\%21-10}{100}){\mathrm{\&Sigma;}}_{b\&Subset;M}\&lsqb;b\&Subset;T_{R,a}\&rsqb;,$

Wherein rand is any random integers, and b is any dough sheet in grid M, and to the layout following constraint bar of example addition Part：

${\mathrm{\&Sigma;}}_{T_{R,a}\&SubsetEqual;T_R}{T}_{R,a}\&GreaterEqual;\frac{\stackrel{\&OverBar;}{N}}{N_T}\&times;0.9,$

WhereinIt is average foreground object number, N contained by every image in image set S_TIt is template number, it is real by calculating layout Example under specifying constraint discretization global optimum layout, obtain repeating objects current grid and constraint under the conditions of most Excellent discrete compact arrangement, i.e. value are 1 layout example T_R,aSet；

Step 2-4, judges whether to receive：The layout example T that the value for obtaining is 1 is solved in statistic procedure 2-3_R,aNumber C, ifThen go to step 3-1；Otherwise, update the value of Q in step 2-2 for Q 1.2, and go to step 2-2.

4. method according to claim 3, it is characterised in that in step 2-1, template T_RTopological structure be defined as follows：

To grid M_RIn all nodes set up by V_BIn all nodes composition communication path and V_IIn all nodes composition most Small spanning tree, wherein V_BIt is grid M_RIn boundary node set, V_IIt is grid M_RIn non-boundary node set, calculate first Each node motion is stored as sequence L to the coordinate direction of next node in communication path and minimum spanning tree_R, L_R=(l₁, l₂,…l_n), l_k∈ { 0,1,2,3,4 }, wherein 1≤k≤n-1, n are grid M_RIn node total number, l_kRepresent in communication path K node v_kWith+1 vertex v of kth_k+1Between spatial relation, its value represents x for 0_k>x_k+1, it is 1 to represent y_k>y_k+1, it is 2 represent x_k≤x_k+1, it is 3 to represent y_k≤y_k+1, wherein x_k、y_kV is represented respectively_kAbscissa and ordinate, x_k+1、y_k+1Represent respectively v_k+1Abscissa and ordinate, while calculate minimum spanning tree in all branch nodes in L_RIn sequence number, be stored as sequence B_R, L_RAnd B_RCollectively constitute repeating objects OImg_RCorresponding template T_RTopological structure.

5. method according to claim 4, it is characterised in that step 3 is comprised the following steps：

Step 3-1, object transformation：Value to being produced in each step 2-3 is 1 layout example T_R,aCorresponding repeating objects OImg_RAngle is carried out for r (OImg_R, rotation and ratio a) are s (OImg_R, a) it is the scaling of ratio, after generation affine transformation Rendered object OImg_rs,a, and determine the drafting center v on painting canvas_c,a；

Step 3-2, hierarchical ranking：The rendered object area that all layout examples are crossed according to corresponding affine transformation is from small to large Sequence, formation sequence Q；

Step 3-3, objects draw：According to the order of Q, after the affine transformation produced in all step 3-1 is drawn on painting canvas Rendered object, wherein for arbitrarily drawn object OImg_rs,a, position is drawn with v_c,aCentered on.

6. method according to claim 5, it is characterised in that in step 3-1, anglec of rotation r is calculated by equation below (OImg_R,a)：

$r({\mathrm{OImg}}_R,a)=\frac{{\mathrm{\&Sigma;}}_{v_{kv,a}\&Element;T_{R,a}}\arctan \left(\frac{y_{v_{kv,a}}-y_{v_{c,a}}}{x_{v_{kv,a}}-x_{v_{c,a}}}\right)-{\mathrm{\&Sigma;}}_{v_{kv}\&Element;T_R}\arctan \left(\frac{y_{v_{kv}}-y_{v_{c,R}}}{x_{v_{kv}}-x_{v_{c,R}}}\right)}{n_{vR}},$

Wherein v_kv,aIt is layout example T_R,aIn grid M_RIn any summit,Respectively v_kv,aHorizontal stroke, ordinate, v_c,aIt is layout example T_R,aThe centre of form,Respectively v_c,aAbscissa and ordinate, v_kvIt is template T_RIn it is any Summit,Respectively v_kvAbscissa and ordinate, v_c,RIt is template T_RThe centre of form, Respectively v_C,RHorizontal stroke Coordinate and ordinate, n_vRIt is template T_RThe node total number that includes of grid.

7. method according to claim 6, it is characterised in that in step 3-1, pantograph ratio s is calculated by equation below (OImg_R)：

$s\left({\mathrm{OImg}}_R\right)=\frac{{\mathrm{\&Sigma;}}_{e_m\&Element;T_{R,a}}{e}_m}{{\mathrm{\&Sigma;}}_{e_m\&Element;T_R}{e}_m},$

Wherein e_mIt is template T_RGrid in any a line.