CN101807304B - Texture synthesis method based on multiplexing - Google Patents

Texture synthesis method based on multiplexing Download PDF

Info

Publication number
CN101807304B
CN101807304B CN2010101318475A CN201010131847A CN101807304B CN 101807304 B CN101807304 B CN 101807304B CN 2010101318475 A CN2010101318475 A CN 2010101318475A CN 201010131847 A CN201010131847 A CN 201010131847A CN 101807304 B CN101807304 B CN 101807304B
Authority
CN
China
Prior art keywords
texture
block
size
texture block
synthetic
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
CN2010101318475A
Other languages
Chinese (zh)
Other versions
CN101807304A (en
Inventor
王文成
陈昕
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Institute of Software of CAS
Original Assignee
Institute of Software of CAS
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Institute of Software of CAS filed Critical Institute of Software of CAS
Priority to CN2010101318475A priority Critical patent/CN101807304B/en
Publication of CN101807304A publication Critical patent/CN101807304A/en
Application granted granted Critical
Publication of CN101807304B publication Critical patent/CN101807304B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Image Generation (AREA)

Abstract

The invention discloses a texture synthesis method based on multiplexing, which comprises the following steps: after analyzing the texture of the sample, generating a small texture block by using a texture block interval distribution synthesis method; selecting a group of new larger texture blocks from the synthesized texture region according to the phase positions, and filling the larger texture blocks into the target texture so as to increase the size of the synthesized texture region; and continuously selecting a new larger texture block from the synthesized texture region through multiplexing to enlarge the synthesized texture region until the target texture is generated. By using the method of the invention, the completed block selection calculation and seaming calculation can be multiplexed, thereby reducing the amount of calculation for block selection and seaming calculation and saving the synthesis time; all the synthesis steps can be realized on CPU, thereby being beneficial to avoiding generating oversize texture due to space limitation; and the invention adopts the multithreading acceleration technology and can acquire higher synthesis velocity.

Description

A kind of based on multiplexing texture synthesis method
Technical field
It is synthetic to the present invention relates to texture, belongs to computerized algorithm, computer graphics techniques, image processing technique, texture generation technique field, is a kind of based on multiplexing texture synthesis method specifically.
Background technology
The texture synthetic technology can generate visually very similar bulk sample according to a fritter sample, can effectively reuse the result of the calculating or the measurement of illumination and color like this, so that generate high-quality drawing result with less expense.This technology has very important in a lot of fields such as sense of reality drafting, virtual reality and uses widely.
The texture synthetic technology mainly is to calculate according to the Markov chain probability model at present, and promptly the color of arbitrary position is to be determined by other COLOR COMPOSITION THROUGH DISTRIBUTION near the certain limit it.And with regard to synthetic order, the texture synthetic technology is divided into serial synthetic method and parallel synthetic method.The each texture block that generates a pixel or comprise some pixels of serial synthetic technology, in general speed is relatively slow.And parallel synthesis technique generates a plurality of pixels or texture block simultaneously, can effectively utilize GPU figures hardware to quicken.At present, synthetic top-quality method is texture optimization method (textureoptimization) (Kwatra V, the Essa I that carries out global optimization computation; Bobick A; Et al.Texture optimization for example-basedsynthesis.ACM Trans.Graph, 2005,24 (3): 795-802); And the fastest method of aggregate velocity is parallel controlled texture synthesis method (parallel controllable texture synthesis) (Lefebvre S; Hoppe H.Parallel controllabletexture synthesis.ACM Trans.Graph., 2005,24 (3): 777-786) and texture block texture synthesis method (Xin Chen spaced apart; Wencheng Wang.Texture synthesis by interspersing patches in a chessboardpattern.In VRCAI ' 09:Proceedings of the 8th International Conference on Virtual RealityContinuum and its Applications in Industry (New York; NY, USA, 2009); ACM, pp.133-138.).Wherein, the latter can synthesize the texture of 1024*1024 pixel in real time, be suitable for the synthetic of large texture, but speed is still limited, and because the restriction in GPU video memory space is unfavorable for generating supersize texture.From the synthetic application demand of texture, it is necessary generating high-quality large texture in real time, and therefore, generate high-quality texture synthetic technology in real time is international hot research content in recent years always.
Summary of the invention
For solving the problems referred to above that exist in the prior art, the present invention proposes a kind of based on multiplexing texture synthesis method, can generate bigger texture fast.
In the present invention, " texture block " refers to the rectangle block with certain size that obtains behind the sample texture piecemeal, and texture block comprises the main part that is positioned at the texture block middle part and is positioned at main part lap on every side.Fig. 1 representes the synoptic diagram of a texture block, in this texture block, texture block be of a size of 9*6 (unit: pixel, down with), main part is of a size of 7*4, lap is that width is 1 straight-flanked ring around the main part.The division of texture block cited below and filling are all carried out according to the size of texture block main part; When promptly dividing texture block; Cut-off rule is two main part sidelines that adjacent texture block is shared, when filling texture block, and two the shared main part of adjacent texture block sidelines.
In the present invention, the definition of " texture phase place " and processing are as follows:
Copy the description of periodic function, after confirming a reference position, with the size of mesh opening that generates texture block sample texture is carried out sizing grid gained texture block when evenly dividing, being called is the texture block of same phase place; The division that reference position difference and gained texture block do not overlap, the texture block that is called out of phase is divided; If the cut-off rule of two kinds of divisions near the time, with regard to claiming these two kinds texture block of dividing gained close phase place is arranged.The texture block of 2 texture phase places that Fig. 2 representes to obtain from sample texture, promptly the texture block shown of solid-line rectangle frame table belongs to a phase place, and the texture block that dotted rectangle is represented belongs to another phase place.
In order to reach technical purpose of the present invention, the present invention adopts following technical scheme:
A) analyzing samples texture finds and can reflect texture block main part size that its textural characteristics of overall importance changes and the size that can be convenient to retrain the lap of selecting texture block;
B) according to a) size of gained, choose suitable texture block size, that is: establish main part and be of a size of a*b, and the width of lap is t that then the size of texture block is (a+t/2) * (b+t/2); On target texture, with the texture of a texture block fritter of synthetic method generation spaced apart, the texture of generation is the global feature and the cyclical variation situation of reflected sample texture preferably, and concrete synthetic method will be introduced in the back;
C) choose bigger texture block size, that is: the main part size becomes (p a* a) * (p b* p wherein b), a, p bBe positive integer, and along with the step c) multiplexing number increases and dull the change greatly, overlapping region width t remains unchanged then; Principles of selected one group new texture block close from the synthetic texture region of step b) according to phase place; Constitute alternative collection M, these new texture block include the last texture block of selected size when synthetic of integer, and the texture block of these sizes has close phase place; And (among the step b) selected phase place is close when initial synthetic; Like this, the texture block among the M just can be spliced each other, and can adapt with synthetic zone;
D) from M, choose texture block arbitrarily, be filled in the target texture, the size of synthetic texture region is enlarged with the mode of linearity, concrete, make its length and width all expand as full-sized twice; Here, employing be the texture block of choosing in the synthetic zone, therefore, reusable is last to select piece and sew application when synthetic;
E) sew up overlapping region between the texture block insert;
F) repeat c)~e) size until synthetic zone reaches the required size of target texture, at this moment, and synthetic the completion.
Wherein, During synthetic initial texture zone; The inventive method has adopted texture block synthetic method spaced apart (Xin Chen, Wencheng Wang.Texture synthesis by interspersing patches in a chessboard pattern.In VRCAI ' 09:Proceedings of the 8th International Conference on Virtual Reality Continuum and itsApplications in Industry (New York, NY; USA; 2009), ACM, pp.133-138.).Its concrete generation step is following:
1) analyze to obtain suitable piece size (with reference to top step a), all texture block that meet this size have constituted set M, with these texture block according to them each self-corresponding texture phase place carry out taxonomic organization, the texture block of the corresponding class of each phase place;
2) to each the texture block Mi among the M, confirming in the every other texture block can be on the four direction up and down of Mi and the texture block of Mi amalgamation, obtain respectively compatibility in abutting connection with set Mi go up, under the Mi, a Mi left side and Mi be right; Such as, with regard to a Mi left side, this set comprises can be in the left side of Mi and the texture block of Mi amalgamation in the texture block of all except that Mi among the M; At this, if two texture block in the difference of overlapping region less than the threshold values of setting, claim that then they are capable assemblings.
When 3) synthesizing a target texture, selected at random texture phase place, and obtain the contiguous several phase places of this phase place, form the suitable texture block set of current synthetic calculating then by the close texture block of these several texture phase places;
4) press the texture block size to the target texture piecemeal; In blank block, insert any in the said suitable texture block set with horizontal and vertical mode (similar chess checkerboard configuration) all spaced apart; Promptly; For each blank block that need fill, all from said suitable texture block set, select one at random;
5), get each blank block corresponding compatibility of texture block all around and fill in abutting connection with the arbitrary texture block in the common factor of set for filling the remaining blank block in back; When occuring simultaneously when empty, just be chosen in these set a maximum texture block of occurrence number and fill.
6) lap between the stitching texture block is to accomplish the generation of a target texture.
In the methods of the invention, the width of size of texture block main part (shape) and overlapping region has material impact for final texture synthetic effect.
The inventive method is further according to document (Wang Yiping; Wang Wencheng; Wu Enhua. the computation optimization that the piece texture is synthetic. computer-aided design (CAD) and graphics journal; 2006,18 (10): method 1502-1507) proposes to decide the size of dividing texture block through the reflection degree that the texture block of different size is divided sample texture periodicity global characteristics.Particularly, make the tolerance of two aspects, promptly the texture of the packets of information capacitive of texture block tolerance and texture block is periodically measured, and with the selection of all reasonable texture block size of these two kinds of metric parameter as division.If two kinds of metric parameter of this under certain size all are reasonable, then this this size can be used as the size of texture block size.The concrete calculation procedure of these two kinds of tolerance is following:
Packets of information capacitive tolerance
A kind of packets of information capacitive of texture block of size is meant the containing reflection degree of the texture block of such size to sample texture information.Its calculation procedure is following:
(1) grey level histogram of calculating sample texture;
(2) take out each texture block of size so successively, calculate its grey level histogram;
(3), and calculate the L between them with the grey level histogram normalization of texture block and sample texture 2Distance.Distance is near more, and then texture block has better reflection to the information of sample texture characteristic of overall importance.
(4) if all the characteristic of overall importance with sample texture is close texture block for the major part under this size (such as more than 90%), the texture block under this size just has good packets of information capacitive so.
Texture is periodically measured
(1) sample texture is divided into bigger grid equably, then to each maybe size piece, one of the picked at random reference block of size like this in each grid;
(2) these reference blocks are carried out classification processing according to the similarity of texture and structural characteristic, promptly relatively whether two reference blocks are similar, if similarly just be classified as one type.Calculating in this similarity is with two overlapping placements of reference block, calculates the L between them then 2Distance.If this distance value shows then that less than the threshold values of a setting these two reference blocks are similar.
(3) number to the reference block of each type averages.If each reference block under a kind of size can both find similar piece, and the mean value of reference block number is higher in these types, and then the piece of this size is just relatively good to the periodically variable reflection of texture information.
Width to the overlapping region; We further according to document (Wang Yiping, Wang Wencheng, Wu Enhua. the computation optimization that the piece texture is synthetic. computer-aided design (CAD) and graphics journal; 2006; 18 (10): method 1502-1507) proposes the width of various overlapping regions is carried out the restrictive tolerance in overlapping region, and decides suitable overlapping region width according to the tolerance result parameter, promptly selects restrictive strong width for use.Its step of metric calculation to a width is following:
(1) sample texture is divided into network equably, each grid all size than texture block is big.In each grid, get a reference zone then respectively, its shape is exactly shape and the size that overlapping region width to be investigated is determined with size.
(2), adopt L to these reference zones 2Distance is as the similarity principle, with they classification.
(3), adopt L to each type reference zone 2Distance is searched for overlapping region similarly as matched rule in sample texture.
Whether similar, and in view of the above these texture block are classified if (4) investigating these pairing texture block in similar overlapping region, to obtain the texture block kind number that such reference zone can be complementary.L is still adopted in similar calculating at this 2Distance is as the similarity measurement criterion.
The texture block kind number that (5) all kinds of reference zones under this width can be complementary averages.Average is more little, and then the binding character of this width is strong more.
In the methods of the invention, step b only generates a texture than the zonule, and texture block is chosen from sample texture; The generation of target texture remainder realizes in subsequent step c~g that then, texture block is chosen among the step c here from synthetic zone, synthetic texture region is to carry out by the end of last step f to finish the texture region that is generated.
In the methods of the invention; Texture block among the step c comprises an integer full-sized texture block; As long as the phase place of the life size texture block that it comprised is close with the phase place of synthesizing the texture block that is adopted in the texture; Can guarantee the coupling between texture block, therefore, significantly reduce the operation of selecting piece and stitching.
In the methods of the invention, for d) in selected texture block carry out certain disturbance, that is: suppose the coordinate of certain selected texture block in synthetic zone for (x, y), then (x+/-δ * p a* a, y+/-δ * p b* b) an optional texture block is replaced original selected piece (x y), like this, under the prerequisite of the basic holistic distribution characteristics that guarantees synthetic result, can be strengthened the randomness of the texture result of generation in the scope; Employed disturbance operator δ adopts different values according to the sample texture difference, and usually, the randomness of sample texture characteristic distribution is strong more, and the value of δ is just big more.The span of δ is in [0,0.25].
In the methods of the invention, step f sews up the lap between the texture block with the mode of sprouting wings.Particularly, emergence refers to the linear interpolation calculating of the pixel of lap position being carried out color, C=C A* w A+ C B* w B, here, C representes to merge the color of back gained, C AAnd C BRepresent respectively from A texture block and B texture block color, w at this same pixel location AAnd w BBe the weights of expression interpolation calculation, be decided by the distance of this pixel to the border, overlapping region.
In said method, institute all accomplishes in CPU in steps.The texture block that adopts among step b synthetic method spaced apart is a parallel synthetic method, and the multiplexing stage (only need phase restriction among the step c~g) between piece and the piece, but also executed in parallel, therefore, above-mentioned two stages multithreading all capable of using quickens.
Compare with prior art, advantage of the present invention and good effect are:
Through multiplexing, constantly utilize synthetic result to generate follow-up texture result, existingly select piece to calculate and sew up and calculate and can be re-used, reduced the calculated amount of selecting piece and stitching, saved generated time; Because the texture piecemeal carries out according to texture cyclical variation characteristic, therefore, the synthetic texture of this method is the overall periodic feature of reflected sample texture well also, and can generate high-quality texture.Though parallel controlled texture synthesis method and texture block synthetic method spaced apart have all obtained higher aggregate velocity, they are all realized on GPU, all can't synthesize when synthetic texture dimensions has in limited time than big or application conditions; And method of the present invention can realize generating bigger texture on CPU, and in addition, the inventive method has adopted the speed technology of multithreading, can obtain higher aggregate velocity.Therefore the present invention can generate the large texture of 2048*2048 pixel in real time, and this to be prior art do not accomplish.
Description of drawings
Fig. 1 representes the texture block synoptic diagram, wherein: 1-main part, 2-lap;
Fig. 2 representes the texture phase place synoptic diagram that texture block is divided, and wherein, the texture block that the solid-line rectangle frame table shows belongs to a phase place, and the texture block that dotted rectangle is represented belongs to another phase place.
Fig. 3 is the synoptic diagram of the inventive method first-order multiplexes, and wherein, 3-representes initial synthetic zone, and 4-representes that 5-representes target texture from synthesizing the new texture block of choosing the result.
Fig. 4 is the multiplexing process flow diagram of the inventive method, wherein: the E-sample texture; P iThe texture block that-each stage is chosen, i=0,1,2; S iThe generation situation of-each phase targets texture, j=0,1,2.
Fig. 5 representes sample texture and target texture instance graph, wherein: the a-sample texture; The target texture that b-the present invention obtains; The target texture that the c-global optimization method obtains; The target texture that the parallel controllable method of d-obtains;
Fig. 6 representes sample texture and target texture instance graph, wherein: the a-sample texture; The target texture that b-the present invention obtains; The target texture that the c-global optimization method obtains;
Fig. 7 representes sample texture and target texture instance graph, wherein: the a-sample texture; The target texture that b-the present invention obtains; The target texture that the c-global optimization method obtains;
Fig. 8 representes sample texture and target texture instance graph, wherein: the a-sample texture; The target texture that b-the present invention obtains; The c-sample texture; The target texture that d-the present invention obtains;
But Fig. 9 is illustrated in and generates the target texture instance graph on the planation surface, wherein: a, the target texture that b-the present invention obtains.
Embodiment
Below in conjunction with accompanying drawing and specific embodiment the present invention is described in further detail:
Embodiment 1
Present embodiment obtains the target texture of 256*256 (shown in Fig. 5 b) through the sample texture (shown in Fig. 5 a) of 128*128.
The entire method flow process is as shown in Figure 4: by choosing P among the sample texture E 0The texture block of size, the synthetic method spaced apart according to texture block generates S 0Shown in the synthetic texture region of meaning; Again from S 0In choose P in the zone of synthetic texture 1The texture block of size is connected with the texture region that has generated, generates S 1Shown in the synthetic texture region of meaning; If also do not reach the target texture size, continue from S 1In choose P in the zone of synthetic texture 2The texture block of size is connected with the texture region that has generated, until reaching the target texture size.
Specifically comprise the following steps:
A) analyzing samples texture finds the texture block main part of its textural characteristics of overall importance of reflection and the size of lap; Texture block main part size is set through following method:
I. set various main part sizes, such as: 2*2,3*3,4*3,6*5,7*2 or the like, can estimate sample texture and do this setting;
Ii. according to packets of information capacitive tolerance various sizes are sorted:
Be the content of example descriptor pardon tolerance below with 4*3, other sizes are all made identical tolerance:
1. divide sample texture (not considering the texture cycle this moment) with 4*3 as the main part size and obtain n texture block;
2. calculate the grey level histogram and the said n texture block grey level histogram separately of sample texture;
3. with the grey level histogram of each texture block and the grey level histogram normalization of sample texture, and calculate the L between them 2Distance;
4. set a threshold value ep, if this L 2Distance is then thought this texture block packets of information capacitive " good " less than ep, otherwise, think " bad ";
5. calculate in this n texture block, the shared ratio of texture block of packets of information capacitive " good " is the metric of this ratio as this size of 4*3.
After each size all accomplished tolerance, various sizes are carried out the front and back ordering according to the height of said ratio, such as 4*3,4*4,5*4 ...
Iii. periodically measure according to texture various sizes sorted:
Be that example is described the content that texture is periodically measured below with 4*3, other sizes are all made identical tolerance:
1. divide sample texture (not considering the texture cycle this moment) with 4*3 as the main part size;
2. sample texture is divided into several bulks, such as, every bulk is of a size of 16*16, and then sample texture (128*128) can be divided into 8 * 8=64 piece;
3. from each piece, randomly draw a texture block, obtain 64 texture block;
4. these 64 texture block are divided into the m class, The classification basis is: in the same classification, and the L between some texture block and all the other texture block 2Distance is less than preset threshold ep2;
5. if the texture block more than 2 is all arranged in each type, and the mean value of the corresponding texture block number of these types is than higher, and then the periodicity metric of this 4*3 size is just high, can be with the periodicity metric parameter value of this mean value as this size.
After each size all accomplished tolerance, various sizes are carried out the priority ordering according to their size of metric parameter value, such as 6*5,4*4,4*3,5*4 ...
Iv. according in step I i and iii, all being positioned at the size of the size in prostatitis as said main part, such as, with 4*4 as concrete size.
The lap size of texture block is set through following method:
I. setting various lap sizes, is 1,2,3 such as width around the main part of 4*4 ... Square loop;
Ii. sample texture is divided into several bulks, such as, every bulk is of a size of 16*16, and then sample texture (128*128) can be divided into 8 * 8=64 piece;
Iii. for each lap size, all make following tolerance:
From each bulk, randomly draw a reference zone piece and obtain 64 reference blocks; Said 64 reference blocks are divided into big type of u according to similarity; To each reference block of big type, in sample texture, find all zones similar, and obtain texture block thus by these range constraints with them; Then, these texture block are carried out similarity classification, to obtain one big type the reference block institute number of texture block class accordingly; At last, the number of the texture block class that all kinds of reference blocks are corresponding averages, and promptly obtains the metric parameter value of the restriction ability of overlapping region under this width.
Iv. from small to large various lap sizes are carried out front and back orderings by above-mentioned metric parameter value, with the size that is positioned at the prostatitis as concrete size.
B) according to a) size of gained; Choose suitable texture block size, and generate an initial texture region according to texture block method spaced apart, (target sizes of this step of setting is 128*128 to size for 130*130; Consider the width 2 of overlapping region, therefore generate 130*130);
The synthetic method that this step is concrete is following:
1) all texture block that meet this size in the sample texture have constituted set M; For any the texture block Mi among the M; Confirming in the every other texture block can be on the four direction up and down of Mi and the texture block of Mi amalgamation, obtain respectively compatibility in abutting connection with set Mi go up, under the Mi, a Mi left side and Mi be right; If M comprises x texture block, the compatibility that then obtains has 4x in abutting connection with set;
2) by the texture phase place M is divided into a plurality of subclass, therefrom selects the close a plurality of subclass of a sub-set or phase place to form a suitable texture block set, to carry out the synthetic calculating of texture;
3) press the texture block size to the target texture piecemeal, in blank block, insert any in the said suitable texture block set with horizontal and vertical mode all spaced apart;
4) for filling the remaining blank block in back; Get each corresponding compatibility of texture block around blank fast and fill in abutting connection with the arbitrary texture block in the common factor of set, that is, and when filling a certain blank block; The Mi that gets its left side texture block is right; The Mi of a right side texture block left side, the Mi of top texture block down with the Mi of below texture block on, get concentrated any one of these four intersection of sets then and fill; If do not occur simultaneously, just select the texture block that occurrence number is maximum in these four set to fill;
5) sew up the lap between the texture block with the mode of sprouting wings.
C) size of expansion texture block, each of the synthetic regional length and width of selection here is half the to be new texture block size, as shown in Figure 3, considers the influence of overlapping region, new texture block size has just become (64+2) * (64+2), i.e. 66*66; To comprise complete life size texture block is standard finds 66*66 in synthetic zone texture block, forms and is suitable for set, and the texture block in this set all has close phase characteristic;
D) at c) in optional 12 texture block in the set that generates, with the mode of imbricate with synthetic regional be connected or join each other be filled in the target texture, can generate and be of a size of 258*258;
E) for d) in selected texture block, add some disturbances, for coordinate be (x, texture block y) in (x+/-2, y+/-2) scope optional texture block replacement texture block (x y), can add certain random effect in the result;
F) sew up the lap between the texture block with the mode of sprouting wings;
G) calculate the piece size that generates, 258*258 has reached the requirement of target texture 256*256, removes the border, and the part of intercepting 256*256 is as generating the result; Generate bigger texture if desired, can get back to step c, next time multiplexing of beginning.
In addition, present embodiment is respectively according to global optimization method (Kwatra V, Essa I; Bobick A; Et al.Textureoptimization for example-based synthesis.ACM Trans.Graph, 2005,24 (3): 795-802.) with parallel controllable method (Lefebvre S; Hoppe H.Parallel controllable texture synthesis.ACM Trans.Graph.; 2005,24 (3): 777-786) obtain target texture, respectively shown in Fig. 5 c and 5d by identical sample texture.
Embodiment 2
Present embodiment through the method identical with embodiment 1 respectively by target texture shown in sample texture composite diagram 6b shown in Fig. 6 a and Fig. 7 a and Fig. 7 b.
Simultaneously, according to global optimization method by target texture shown in the 6c of sample texture composite diagram shown in Fig. 6 a, and according to parallel controllable method by target texture shown in the 7c of sample texture composite diagram shown in Fig. 7 a.
Visible by Fig. 5-7; The inventive method can generate high-quality texture; Can be suitable with at present synthetic top-quality global optimization method; Even can keep the structured message of texture better, can destroy the more inner structurized unit of texture like the global optimization method among Fig. 6, and the present invention can not yet.
Embodiment 3
This instance is through having generated the target texture of Fig. 8 b, Fig. 8 d, Fig. 9 a, Fig. 9 b with instance 1 identical method, and the size of the target texture here is 512*512.
By Fig. 8,9 visible, the inventive method can better quality generate polytype texture, but in application such as planation surface, all can be well used.
Embodiment 4
Present embodiment on Dell Optiplex 755 microcomputers by the target texture of the synthetic multiple size of same sample texture.
This PC is furnished with one
Figure GSA00000065196000101
Core TM2 Duo E6550 2.33GHz CPU, 2G internal memory and a NVIDIAGeforce 8600GTS GPU.
Experiment shows that the present invention is the texture of Synthetic 2 048*2048 pixel in real time, and is well suited for generating large-sized texture, sees table 1.
According to people (Han C such as Han C; Risser E, Ramamoorthi R, Grinspun E.Multiscale texture synthesis.ACM Trans.Grap.; 2008; 27 (3), Article No.51) work, the treatment effeciency of going up parallel controlled texture method at the higher machine (GeForce 8800 GTXGPU) of configuration is 15.3M pixel/second.According to this treatment effeciency, we have conversed its generated time for each size texture, in table 1, list, and can find out, the inventive method has higher speed than parallel controllable method, is well suited for generating the large scale texture.Though method (the Xin Chen of interval cloth; Wencheng Wang.Texturesynthesis by interspersing patches in a chessboard pattern.In VRCAI ' 09:Proceedings of the 8thInternational Conference on Virtual Reality Continuum and its Applications in Industry (New York, NY, USA; 2009); ACM pp.133-138.) also is convenient to the generation of large texture very much, but speed wants slow than the inventive method; And receive the GPU restriction can't generate very big texture; And the inventive method has combined its advantage effectively, and in CPU, calculates, and can generate the texture of 2048*2048 pixel in real time.
Synthetic counting yield of table 1 present embodiment texture and comparison:
Texture dimensions (pixel) The new method generated time The generated time of interval cloth The generated time (conversion) of parallel controllable method
256*256 0.68 3.51 4.08
512*512 1.39 4.16 16.34
1024*1024 3.81 6.94 65.36
2048*2048 13.47 17.89 261.44

Claims (5)

1. one kind based on multiplexing texture synthesis method, comprises the following steps:
A) analyzing samples texture is confirmed the main part size of ability reflected sample texture textural characteristics of overall importance and is convenient to the lap size that the texture block constraint is chosen;
B) according to said main part size and lap size, choose the texture block that is of a size of (a+t/2) * (b+t/2), the synthetic method spaced apart with texture block generates a texture on target texture; Said main part is of a size of a*b, and the width of lap is t;
C) from target texture the synthetic texture region according to one group of new texture block of the close principles of selected of phase place, constitute alternative collection M, new texture block main part size becomes (p a* a) * (p b* b), t is constant for the lap width, wherein p a, p bBe positive integer, and along with the step c) multiplexing number increases and dull the change greatly;
D) from alternative collection M, choose texture block, be filled in the target texture, the size of synthetic texture region is enlarged with the mode of linearity;
E) sew up overlapping region between the texture block insert;
F) repeat c)~e) until synthesizing the size that regional size reaches target texture, then accomplish synthetic.
2. texture synthesis method as claimed in claim 1 is characterized in that, synthetic texture region is to carry out by the end of last step e to finish the texture region that is generated in the said step c).
3. texture synthesis method as claimed in claim 1 is characterized in that, said step d) is chosen texture block at random from alternative collection M.
4. texture synthesis method as claimed in claim 3 is characterized in that, carries out certain disturbance when choosing texture block; To increase the randomness of synthetic texture; The concrete operations of disturbance are: suppose the coordinate of certain selected texture block in synthetic zone for (x, y), then at (x ± δ * p a* a, y ± δ * p b* b) optional texture block is replaced selected texture block (x, y), the span of disturbance operator δ is in [0,0.25] in the scope.
5. texture synthesis method as claimed in claim 1 is characterized in that, said step a)~f) adopted the accelerated method of multithreading on CPU, to realize.
CN2010101318475A 2010-03-23 2010-03-23 Texture synthesis method based on multiplexing Expired - Fee Related CN101807304B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010101318475A CN101807304B (en) 2010-03-23 2010-03-23 Texture synthesis method based on multiplexing

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010101318475A CN101807304B (en) 2010-03-23 2010-03-23 Texture synthesis method based on multiplexing

Publications (2)

Publication Number Publication Date
CN101807304A CN101807304A (en) 2010-08-18
CN101807304B true CN101807304B (en) 2012-02-22

Family

ID=42609086

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010101318475A Expired - Fee Related CN101807304B (en) 2010-03-23 2010-03-23 Texture synthesis method based on multiplexing

Country Status (1)

Country Link
CN (1) CN101807304B (en)

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101937567B (en) * 2010-09-28 2012-01-18 中国科学院软件研究所 Method for extracting main texture simply and quickly
CN103473750B (en) * 2013-08-02 2017-12-15 毕胜 The method for rendering fabric border anastomosing and splicing
CN103646428B (en) * 2013-11-29 2016-06-01 华东师范大学 The synthetic method of a kind of 3D fluid scene
CN105957137B (en) * 2016-04-20 2018-09-21 上海交通大学 Three-dimensional colour model image processing method based on texture mapping
CN107610197A (en) * 2017-09-18 2018-01-19 浙江科澜信息技术有限公司 A kind of texture merging method and system
CN112417570A (en) * 2020-12-03 2021-02-26 杭州群核信息技术有限公司 Continuous grain brick paving design method and system and computer equipment

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR100395075B1 (en) * 2001-05-18 2003-08-19 최종수 Method for constructing of 2D image mosaics using quasi feature point
CN1256707C (en) * 2004-05-09 2006-05-17 北京航空航天大学 Grain synthesizing method based on multiple master drawings
CN101441775B (en) * 2008-12-09 2010-09-08 中国科学院软件研究所 Texture synthesis method based on matching compatibility
CN101635047A (en) * 2009-03-25 2010-01-27 湖南大学 Texture synthesis and image repair method based on wavelet transformation
CN101620737B (en) * 2009-08-07 2011-09-21 中国科学院软件研究所 Texture synthesis method based on interval distribution of texture blocks

Also Published As

Publication number Publication date
CN101807304A (en) 2010-08-18

Similar Documents

Publication Publication Date Title
CN101807304B (en) Texture synthesis method based on multiplexing
CN103679627B (en) Tile based computer graphics
CN102822870B (en) The process of 3D computer graphics data on multiple shadowed engines
CN102792302A (en) Method and apparatus for creating surface processing data
CN103088515A (en) Method suitable for weaving colored jacquard fabric with large texture thread number
CN103514625A (en) Three-dimensional reconstruction method based on multi-view self-adaptation characteristic registration
CN110039186A (en) A kind of surface model paths planning method applied to laser surface ablation
CN103415851A (en) Method and device for creating surface treatment data
CN102425024B (en) Method for designing jacquard fabric with black-and-white emulation effect by utilizing colored patterns
CN102750405A (en) Computer simulation method of spun-dyed fancy fabrics
CN103456031B (en) A kind of new method of area image interpolation
CN102547338A (en) DIBR (Depth Image Based Rendering) system suitable for 3D (Three-Dimensional) television
CN104001676A (en) Automatic sorting method and production line for large-scale multi-variety mixed-match underwear
CN105574914A (en) Manufacturing device and manufacturing method of 3D dynamic scene
CN107369190B (en) Rubble-based thermodynamic diagram accelerated rendering method
CN204148152U (en) Large-scale multi items is mixed takes underwear automated sorting production line
CN101620737B (en) Texture synthesis method based on interval distribution of texture blocks
CN101441775B (en) Texture synthesis method based on matching compatibility
CN103914825A (en) Three-dimensional model texture coloring method based on image segmentation
CN114386295A (en) Textile computer simulation method based on color separation and color change of colored spun yarns
CN105256435A (en) Texture structure design method for ink painting style fabric
CN108389154B (en) Method for implementing cutting area segmentation technology for parallel drawing in GPU
CN103279973A (en) Three-dimensional image matching system based on mixing and parallel
CN106331678A (en) Hardware-implemented floating-point type viewpoint processing method in glasses-free 3D display system
CN102724528B (en) Depth map generation device

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20120222