CN104268580B - A kind of class caricature laying out images management method based on scene classification - Google Patents
A kind of class caricature laying out images management method based on scene classification Download PDFInfo
- Publication number
- CN104268580B CN104268580B CN201410546275.5A CN201410546275A CN104268580B CN 104268580 B CN104268580 B CN 104268580B CN 201410546275 A CN201410546275 A CN 201410546275A CN 104268580 B CN104268580 B CN 104268580B
- Authority
- CN
- China
- Prior art keywords
- mrow
- msub
- image
- layer
- template
- 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
Links
- 238000007726 management method Methods 0.000 title claims abstract description 22
- 238000000034 method Methods 0.000 claims abstract description 50
- 238000013517 stratification Methods 0.000 claims abstract description 9
- 230000008520 organization Effects 0.000 claims abstract description 7
- 238000005457 optimization Methods 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 3
- 238000009415 formwork Methods 0.000 claims description 2
- 238000005316 response function Methods 0.000 claims 1
- 239000010410 layer Substances 0.000 description 72
- 230000008569 process Effects 0.000 description 15
- 230000006870 function Effects 0.000 description 8
- 238000004458 analytical method Methods 0.000 description 5
- 230000011218 segmentation Effects 0.000 description 5
- 230000008859 change Effects 0.000 description 4
- 239000002245 particle Substances 0.000 description 4
- 238000000926 separation method Methods 0.000 description 4
- 238000010276 construction Methods 0.000 description 3
- 239000011229 interlayer Substances 0.000 description 3
- 238000003909 pattern recognition Methods 0.000 description 3
- 235000007926 Craterellus fallax Nutrition 0.000 description 2
- 240000007175 Datura inoxia Species 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000033228 biological regulation Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 239000004744 fabric Substances 0.000 description 2
- 201000011243 gastrointestinal stromal tumor Diseases 0.000 description 2
- 238000004451 qualitative analysis Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 241000556720 Manga Species 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000001747 exhibiting effect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000012856 packing Methods 0.000 description 1
- 238000010422 painting Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000007493 shaping process Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 239000013589 supplement Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a kind of class caricature laying out images management method based on scene classification, comprise the following steps:Step one, ask for treating the characteristics of image of each image in organization chart picture set;Step 2, quadtrees is built using the numerical distance between said features;Step 3, passes through tried to achieve quadtrees and builds stratification classification tree;Step 4, representative image is asked for using the classification degree between image in classification tree;Step 5, is ranked up according to classification degree between representative image to representative image;Step 6, pieces representative image quantity and the caricature rule of presence together according to selected, dynamic template storehouse is built using character string enumeration methodology;Step 7, matching template is searched for according to the maximum principle of information content that image after sequence is presented in dynamic template storehouse, and optimizes layout, obtains final splicing result.
Description
Technical field
The present invention relates to a kind of image management method, belong to Computer Image Processing and field of Computer Graphics, specifically
Say it is a kind of class caricature laying out images management method based on scene classification.
Background technology
With the arrival in digital photography epoch, digital picture has become one of Digital Media the abundantest, Ren Menke
To shoot all kinds of photos by various mobile devices whenever and wherever possible.Usual people like creating by theme all kinds of photos of gained are shot
Different files are built to be preserved, but be due to may be comprising number with hundred under the huge of number of pictures, each file
The image of meter, will be a thing extremely wasted time and energy as that need to browse the photograph album or search specific some photos.Therefore to this
The image of a little substantial amounts carries out rational tissue and clearly shown, is beneficial to the present invention and photograph album is efficiently managed
Reason, such as document 1:Wang J.D.,Quan L.,Sun J.,et al.Picture collage.Proceedings of the
IEEE Computer Society Conference on Computer Vision and Pattern
Recognition.2006,347-354. with document 2:Philipp Sandhaus,Mohammad Rabbath,Susanne
Boll.Employing aesthetic principles for automatic photo book layout.In MMM'11
Proceedings of the 17th international conference on Advances in multimedia
Modeling-Volume Part I Pages 84-95 are introduced.
In recent years, it is principally motivated in addressing two key issues --- image in photograph album management for photograph album management study person
Scene classification problem and image shows problem.
For image scene classification degree problem, researcher presses generally according to the different characteristic resulted in image
Different themes are classified to image.The similar journey of numerical distance between different levels feature is relied primarily in assorting process
Degree such as document 3:A.Vailaya,M.Figueiredo,A.Jain,H.J.Zhang.Content-based hierarchical
classification of vacation images.In Proc.Of IEEE International Conference on
Multimedia Computing and Sytems(ICMCS),1999,1:518-523.;Document 4:Lowe
D.G.Distinctive image features from scale-invariant key points[J]
.International Journal of Computer Vision,2004,60(2):91-110. with document 5:
Mikolajczyk K.,C.Schmid.A performance evaluation of local descriptors[J].IEEE
Transactions on Pattern Analysis and Machine Intelligence,2005:It 1615-1630. is situated between
Continue, using clustering method such as document 6:Bishop C.M.,Springerlink.Pattern recognition and
machine learning[M].Vol.4:Springer New York.2006:With document 7:Frey B.J.,Dueck
D.Clustering by passing messages beteen data points.Science 315(Feb 2007),
972-976. introduce.Or the method such as document 8 of study:Vailaya A,Figueiredo M,Jain A.Image
classification for content-based indexing.IEEE Transactions on Image
Proeessing,2001,10:117-129. with document 9:Fan J,Gao Y,Luo H.Statistical modeling and
concePtualization of natural images.Pattem Reeognition,2005,38:865-885. introduce.
Image is classified, but these are not fully reliable to the data distance for differentiating classification, and when comparing, two height are not similar
Image when will fail, therefore generation the problems such as can cause to classify inaccurate.
For image shows problem, it is a kind of photograph album way to manage popular recently to piece together, and it can be by one group of figure
As combined and spliced into piece image or a few width images, realize that the concentration to this group of image is summarized.Being based primarily upon this kind of method more
Spatial Rules set up corresponding layout, using different style and technology realize that the combination of image is pieced together such as document 1:Wang
J.D.,Quan L.,Sun J.,et al.Picture collage.Proceedings of the IEEE Computer
Society Conference on Computer Vision and Pattern Recognition.2006,347-354. and
Document 10:Y.Yang,Y.Wei,C Liu,Q.Peng,and Y.Matsushita.An Improved Belief
Propagation Method ofr Dynamic Collage.The Visual Computer,Vol.25,No.5-7,431-
439.2009 with document 11:Zongqiao Yu,Lin Lu,Yanwen Guo,Rongfei Fan,Mingming Liu,et
al.Content-aware photo collage using circle packing.IEEE Transactions on
Visualzation and compter graphics.Vol.20, No.2,182-194.2014 introductions.But this kind of method is more
Be to start with from image shows angle, it is considered to multiple image splices in photograph album compact sense and visual attraction, and seldom close
Note relation that may be present between image.Nearest Zhang et al. is such as document 12:Lei Zhang and Hua
Huang.Hierarchical Narrative Collage For Digital Photo Album.Computer
Graphics Forum Volume 31, Issue 7, pages 2173-2181, September2012 carried based on event
Stratification piece method together, can in chronological order by event mode in photograph album image carry out tissue, can be very good
The function of being browsed into image by fact retrieval, but because this method need to be drawn by the temporal information of same equipment to image
Point, therefore with significant limitation, after being shot using multiple equipment to Same Scene event, this method would become hard to pair
Image carries out tissue.
In addition, for image shows problem, also thering are some investigators to carry out image exhibition by the way of class caricature layout
Show such as document 12:M.Wang,R.C.Hong,X.T.Yuan,S.C.Yan,T.S.Chua.,Movie2Comics:Towards a
Lively Video Content Presentation.IEEE Transactions on Multimedia.Vol.14,
No.3.2012.858-870. with document 13:J.Calic,D.P.Gibson,and N.Campell,“Efficient layout
of comic-like video summaries,”IEEE Trans.Circuits Syst.Video Technol.,
Vol.17, no.7, pp.931-936, Jul.2007. are introduced.Not only picture is succinct for this kind of methods of exhibiting, and due to caricature
Generally there is certain reading order, so being conducive to the information such as retention time sequence or collating sequence.But being adopted existing method more
Realize that class caricature is laid out with fixed several templates, because template number is limited, so easily there is layout in layout process
Form is single, the problems such as typesetting aesthetic property is poor.Although have research can the method based on study it is corresponding from image sequence generation
Caricature layout such as document 14:Cao Y,Chan A B,Lau R W H.Automatic stylistic manga layout
[J].ACM Transactions on Graphics(TOG),2012,31(6):141. are introduced, and are had on the pattern of layout
Very big improvement, but need substantial amounts of man-machine interactively to mark in generating process, and various learning methods are optimized, this
Make that this method amount of calculation is excessive, process is excessively complicated.
The content of the invention
Goal of the invention:The technical problems to be solved by the invention be the scene for making user in image fast and accurately
Finding target image, there is provided a kind of class caricature laying out images management method based on scene classification.
Technical scheme:The invention discloses the class caricature laying out images management method based on scene classification, including following step
Suddenly:
Step one, ask for treating each characteristics of image in organization chart picture set;
Step 2, quadtrees is built using the numerical distance between these features;
Step 3, passes through tried to achieve quadtrees and builds stratification classification tree;
Step 4, representative image is asked for using the classification degree concept between image in classification tree;
Step 5, is ranked up according to classification degree situation between representative image to it;
Step 6, pieces representative image quantity and the caricature rule of presence together, using character string enumeration methodology according to selected
Build dynamic template storehouse;
Step 7, matching mould is searched for according to the maximum principle of information content that image after sequence is presented in dynamic template storehouse
Plate, and layout is optimized, obtain final splicing result.
Step one characteristics of image of the present invention is asked for have selected 3 kinds of the more commonly used features participation calculating.Main choosing of the invention
Color, the SIFT shape facilities such as document 4 of bottom are selected:Lowe D.G.Distinctive image features from
scale-invariant key points[J].International Journal of Computer Vision,2004,
60(2):91-110. with document 5:Mikolajczyk K.,C.Schmid.A performance evaluation of
local descriptors[J].IEEE Transactions on Pattern Analysis and Machine
Intelligence,2005:1615-1630. is introduced, and middle level be used for the GIST features such as document of global scene is described
15:Oliva A.,Torralba A.Modeling the shape of the scene:a holistic
Representation of the spatial envelope.Int.J.Comput.Vison42.3,145-175. are introduced.
The main cause for selecting these features is that these features are often used in the research of scene classification by scholar, and is easy to meter
Calculate.
In step 2 of the present invention after the image in the photograph album using step one method to input asks for features described above, this
Invention can use document 16 based on these features:Shi-Sheng Huang,Ariel Shamir,Chao-Hui Shen,Hao
Zhang,Alla Sheffer,Shi-Min Hu,Daniel Cohen-Or.Qualitative Organization of
Collections of Shapes via Quartet Analysis.ACM Transactions on Graphics
(Proceedings of SIGGRAPH2013), the method that 32 (4) are introduced builds corresponding quadtrees set.Constructed
One quadtrees includes four images, forms two pairs, the spectral discrimination of wherein each pair is similar, is sentenced from different pairs of image
Surely it is dissimilar..
Classification tree constructed in step 3 is a unrooted tree in step 3 of the present invention, its can farthest keep by
The topological relation of image similarity degree embedded by quadtrees, each leaf node of tree represents an image, non-leaf nodes
Represent scene classification situation.If leaf node shares same father node, then prove that these images belong to Same Scene, and
If two non-leaf nodes share same father node, then prove that the scene representated by it has certain similarity degree, because
This can carry out stratification tissue to image by the classification tree by scene.
Analyzed by step 3 of the present invention present invention understands that, a field minimum of the non-leaf nodes representative image of the bottom
Subordinate relation between scape, node represents the similarity degree between scene.Therefore the present invention is carried in step 4 of the present invention by scene
It when taking representative image, can be extracted by different grain size, you can scene size is divided according to hierarchical relationship, extract corresponding generation
Table image.The present invention defines its hierarchical relationship by the separating degree between image, and the value can also embody scene
Similarity degree between class.Separating degree between so-called image refer in a classification tree image to other images to be undergone it is each most
The beeline of short path, as leaf node.This specific calculating process of clearly demarcated step 4 is as follows:
Step (41):From any leaf node, using the separated degree of BFS less than or equal to threshold value
Other leaf nodes, and these nodes are rejected from classification tree, constitute one group of scene image;
Step (42):Repeat said process, until not stopped search in classification tree when comprising image, now remaining figure
As being exactly representative image.
Step 5 of the present invention wishes to select the image of otherness minimum in class as the representative image of the group, and the present invention passes through
Every image is asked for the separating degree sum of other images in the group to define otherness in class, and based on this selection representative graph
Picture.Detailed process is as follows:
Step (51):Ask for the separating degree sum of every image in group;
Step (52):It is ranked up from small to large ord;
Step (53):The minimum image of selected value is (of the invention if there is the equal image of multiple values for the representative image of the group
Representative image of the piece image as the group will be randomly choosed).
Step 6 of the present invention pieces representative image quantity and the caricature rule of presence together selected by, is enumerated using character string
Method builds dynamic template storehouse.It is known that one page caricature layout is often constituted by several layers, wherein every layer by list not of uniform size
First lattice into.In order to there is abundant presentation mode, the present invention wishes that the template of polymorphic type is available for user to select.Because image is
In order, layering is placed, it is possible to generate original template using the method enumerated.Further, since caricature is laid out
During there are many heuristic rules to follow, so the generation of redundancy template can also be reduced by these rules.This
The specific calculating process of inventive step six is as follows:
Step (61):Space will be pieced together using heuristic rule and carry out layered shaping, every page of splicing space can only at most be divided
4 layers are segmented into, every layer can only at most be divided into 3 cells;
Step (62):Put and obtained by enumeration methodology for every layer of each unit case.
Step (63):Gained content will be enumerated to be indicated by character string forms, and every page of spelling is divided according to character string
Space is connect, final corresponding templates are obtained.
Step 7 of the present invention is searched for according to the maximum principle of information content that image after sequence is presented in dynamic template storehouse
With template, and layout is optimized, the preceding top-k matched the most is mainly searched in dynamic template storehouse by defining penalty
Individual template (value that k is determined according to user's request).Specially:Representative image is long as template to maximize capping unit lattice
After wide equal proportion scaling, the sequence representative image after scaling is put into each unit lattice in center alignment mode, due to representativeness
Image may be different with the size of each unit lattice, can produce gap, covering and difference in height phenomenon, and adjacent unit compartment is produced
Above-mentioned phenomenon be defined as local restriction relation;And may be one due to the larger spilling cell of representative area between adjacent layer
Part has been covered in next layer, and global restriction relation is covered as between the such adjacent layer of definition;According to the local and overall situation about
The beam relation present invention is each template TmDefining penalty is:
Wherein:
Penalty(Tm) represent template TmPenalty, α, β is respectively the weight of local restriction and global restriction, and N is
Min-max normalization factors, AREA is the size of caricature template, LiRepresent template TmI-th layer, NL represents template TmLayer
Number, Eo(Li) it is i-th layer of area coverage, Eg(Li) it is i-th layer of gap degree, Eh(Li) it is i-th layer of the degree of balance;Ii,jRepresent
It is put into the presentation graphics of i-th layer of j-th of cell, Ii,j+1Expression is put into the presentation graphics of+1 cell of i-th layer of jth;
Area(Ii,j) it is the presentation graphics area for being put into i j-th of cell of layer, Area (Ii,j+1) it is to be put into+1 unit of i layers of jth
The area of the presentation graphics of lattice;The number for the cell that J is included for i-th layer of the template, Area (Li) it is i-th layer of face
Product;Height(Ii,j) it is the picture altitude for being currently put into i-th layer of j-th of cell, Height (Ii,j+1) it is currently to be put into i layers
The picture altitude of+1 cell of jth;D(Li,Li+1) represent the coverage of i-th layer and i+1 layer, Area (Ii+1,q) it is current
The image area of i+1 q-th of cell of layer is put into, Q represents the cell quantity of i+1 layer;∩ operates to hand over, and represents two areas
Domain area intersects situation;∪ is and operated, and represents Two Areas area combination situation.
It is that k template retains representative area as object definition fitness function to maximize in step 6 of the present invention, and
Take optimization method to optimize fitness function so as to generate the caricature layout that formwork style is abundant, be specially:Select k
After individual template, the presentation graphics after scaling, which is put into after corresponding cell, can still have the phenomenons such as covering, gap, and these are existing
As that will cause the cutting to representative area and supplement, this will all destroy the integrality of presentation graphics.In order to maximize guarantor
Representative area is stayed, while being each template definition optimization aim in k template to generate the template that pattern is abundant
For:
Wherein, TmFor current template;NL template TmThe maximum number of plies, NL ∈ (1...4);J ∈ (1...3) are every layer of institute
Comprising cell number;Pi,jRepresent that the cell is in template TmI-th layer in j-th of position;Ii,jFor by scaling
Representative image afterwards is put into the rectangle after i-th layer of j-th of cell;Ratio (x) (works as cell for the length-width ratio of corresponding region
When deforming upon, the value is the length-width ratio of minimum extraneous rectangle).The present invention optimizes energy function Y using particle cluster algorithm
(Tm), final layout result is recommended into user.
Beneficial effect:The image management method of present invention advantage compared with existing image management method is:First, it is sharp
Rational classification has been carried out to the scene in photograph album with the qualitative analysis image organization method based on quadtrees, scene can have been entered
Row stratification tissue, so that user can be classified according to scene, and quickly find target according to scene, and is passed
The computer picture management method of system, it is impossible to realize the technical functionality.Secondly, design realizes the class based on dynamic template storehouse
The automatic of caricature layout pieces method together, and the representative image in stratification scene can be carried out more succinct to piece displaying together.
Brief description of the drawings
The present invention is done with reference to the accompanying drawings and detailed description and further illustrated, of the invention is above-mentioned
And/or otherwise advantage will become apparent.
Fig. 1 is our bright broad flow diagram.
Fig. 2 is that quadtrees of the present invention builds explanation figure.
Fig. 3 is classification tree situation of the present invention for one group of data.
Fig. 4 is that representative image of the present invention is extracted and sequencer procedure.
Fig. 5 is that the present invention enumerates gained character string situation corresponding with template.
Fig. 6 is layout optimization process of the present invention.
Embodiment:
The present invention is described in further detail with reference to the drawings and specific embodiments:
As shown in figure 1, the present invention comprises the following steps:
Step one, ask for treating each characteristics of image in organization chart picture set;
Step 2, quadtrees is built using the numerical distance between these features;
Step 3, passes through tried to achieve quadtrees and builds stratification classification tree;
Step 4, representative image is asked for using the classification degree concept between image in classification tree;
Step 5, is ranked up according to classification degree situation between representative image to it;
Step 6, pieces representative image quantity and the caricature rule of presence together, using character string enumeration methodology according to selected
Build dynamic template storehouse;
Step 7, matching mould is searched for according to the maximum principle of information content that image after sequence is presented in dynamic template storehouse
Plate, and layout is optimized, obtain final splicing result.
In the present invention, to ask for process as shown in Figure 2 for quadtrees specific.First, to tetra- figures of any A, B, C, D in photograph album
As one Connected undigraph of construction, as shown in a in Fig. 2, wherein the value of each edge is defined in one image of each node on behalf, figure
For the distance of two nodes (it is reciprocal that the present invention is defined as similarity of two images under same feature constraint);Then, according to value
Size is ranked up to 6 sides, first three maximum sides of deletion value, schemes no longer to connect after such as deleting, then four images will
Quadtrees can not be constituted, otherwise, it is d3 to take the maximum side of distance, and remaining two sides are d1, d2, as shown in b in Fig. 3;Finally,
The pair of quadtrees, such as d3/d1 are built by calculating the ratio between d3 and d1 and d2>R, d3/d2>(wherein R is regulation to R
The parameter of four-tuple quantity), as shown in c in Fig. 2, then four images can build a quadtrees, and side is the two of d1
One pair of individual node formation, side forms another pair for d2 two nodes, as shown in d in Fig. 2.Such as it is unsatisfactory for above-mentioned
Condition four images can not build a quadtrees.
For follow-up rapid build classification tree, the present invention also needs to control the generation quantity of quadtrees, mainly passes through regulation
Threshold value R works to complete this.When the setting of R values is smaller, the four-tuple quantity tried to achieve will be excessive, and this will the increase later stage point
The amount of calculation that class tree is formed;When R values setting than it is larger when, required four-tuple quantity will be very few, and this will reduce its reliability, shadow
The generation of later stage classification tree is rung, so R values will rationally be set.60 multiple images are directed to by experiment, under color characteristic,
When R values are set to 2.4 by the present invention, more than 8100 quadtrees can be constructed;Under SIFT feature, R values are set to 3.2 by the present invention,
More than 6400 quadtrees can be constructed;Under GIST features, R values are set to 2.2 by the present invention, can build 8700 quadtreeses.Cause
This present invention will participate in the structure of subsequent classification tree using more than 23200 quadtrees.
In the present invention, after quadtrees has been built, it can continue to use document 16:Shi-Sheng Huang,Ariel
Shamir,Chao-Hui Shen,Hao Zhang,Alla Sheffer,Shi-Min Hu,Daniel Cohen-
Or.Qualitative Organization of Collections of Shapes via Quartet Analysis.ACM
Introduce method to build point in Transactions on Graphics (Proceedings of SIGGRAPH 2013), 32 (4)
Class tree, and then complete the management on levels to inputting photograph album, that is, complete scene classification.The classification tree is a unrooted tree, its energy
Farthest keep as the topological relation of the image similarity degree embedded by quadtrees, each leaf node of tree represents one
Image, non-leaf nodes is to represent scene classification situation, if leaf node shares same father node, then prove these images
Belong to Same Scene, and if two non-leaf nodes share same father node, then prove that the scene representated by it has one
Fixed similarity degree, therefore stratification tissue can be carried out by scene to image by the classification tree.In addition, step 2 is tried to achieve
Each quaternary tree construction can be regarded as a subtree of classification tree, i.e., each quadtrees relation can be searched in tree
Arrive.For one group of image as shown in a in Fig. 3, the classification tree constructed by the present invention is as shown in b in Fig. 3, it can be seen that the classification tree
Can be to image by scene progress Rational Classification.
In the present invention, show, need to select to represent by scene classification situation on the basis of classification tree to carry out summary formula
Property image is pieced together.Known by above analyzing, a minimum scene of the non-leaf nodes representative image of the bottom, node
Between subordinate relation represent similarity degree between scene.Therefore the present invention, can be by not when extracting representative image by scene
One-size is extracted, you can divide scene size according to hierarchical relationship, extracts corresponding representative image.The present invention passes through figure
Separating degree as between defines its hierarchical relationship, and the value can also embody similarity degree between the class of scene.So-called figure
Separating degree as between refers to that an image is to other images each shortest path to be undergone in classification tree, for b institutes in Fig. 3
Separating degree situation between the classification tree tried to achieve, each image as shown in a in Fig. 4, such as in Fig. 3 classification trees marked as 24,25 two
Width image, therefrom an image to an other image needed on classification tree by shortest path be 2, therefore between them
Separating degree be 2.
After separating degree between image is asked for from classification tree, just scene can be entered by different grain size according to user's request
Row is divided.Specific calculating process is as follows:The present invention is separated using BFS first from any leaf node
Degree is less than or equal to dthreshOther leaf nodes, and these nodes are rejected from classification tree, constitute one group of scene image;So
Said process is repeated afterwards, until not stopped search in classification tree when comprising image.By aforesaid operations, the present invention can be by photograph album
In the granularity specified by user of image be divided into multiple series of images, as shown in b in Fig. 4, when choosing dthresh, can be by Fig. 3 when=2
Image be divided into 8 groups of images, every group of image will represent a scene.It should be noted that dthreshThe selection of value, the size of the value by
User is determined, and it determines the segmentation granularity of scene, works as dthreshWhen value selection is smaller, the scene quantity of segmentation is relative
Can be relatively more, work as dthreshValue selection than it is larger when, the scene quantity of segmentation is relative can be fewer.Normal conditions such as image volume is very
Greatly, and each when the leaf node amount that bottom node is included is too small, the present invention need to be by dthreshIt is larger that value is set, so
It ensure that selected representative image quantity will not be excessive, be easy to the follow-up displaying of photograph album.
After being grouped to image, the present invention needs to select a representative image to participate in follow-up class caricature layout spelling in each group
Patch, for one group of image, the present invention wishes the image of otherness minimum in selection class as the representative image of the group, and the present invention is logical
The separating degree sum for asking for every image to other images in the group is crossed to define otherness in class, and representative graph is chosen based on this
Picture.Detailed process is as follows:The present invention asks for the separating degree sum of every image in group first, then carries out from small to large ord
Sequence, for the representative image of the group, (if there is the equal image of multiple values, the present invention will be selected the minimum image of last selected value at random
Piece image is selected as the representative image of the group).The image that every group is known with yellow collimation mark in b in such as Fig. 4, as using present invention side
The representative image that method selection is obtained.
Further, since selected representative image in former classification tree have certain topological relation, i.e. representative image that
There is corresponding separation angle value between this, represent the similarity degree between them.In order to preferably keep this in image shows
A little relations are so as to follow-up quick-searching, and the small representative image of separation angle value to the greatest extent may be used during displaying between the present invention wishes
Can be close, also precisely in order to keeping this relation, the present invention have selected the class caricature layout type pair with level, order information
Image is shown.Therefore before the layout splicing displaying of class caricature is carried out, the present invention is also needed to representative image by between them points
It is ranked up from degree situation.Specific sequencer procedure is as follows:The present invention in representative image randomly choose an image as need spell
First image connect;Ask for the separation angle value between other representative images and the image;By separation angle value from small to large order
(if there is the equal image of multiple values, the present invention will carry out randomly ordered to these images) is ranked up, final sequence is obtained
As a result the input picture for splicing displaying will be laid out as follow-up class caricature.For the selected representative images of b in Fig. 4, final row
Sequence result is as shown in c in Fig. 4, it can be seen that scene relatively similar image sorting position closer to.
Because the space of every page of caricature layout is limited, in order to which bandwagon effect is more aesthetically pleasing, every page of present invention setting is overflow
5-8 images are presented in painting canvas office.For the sequence representative image that inputs of the present invention, the present invention will in order to they carry out with
Machine is grouped, and every group of amount of images is maintained between 5-8, the input calculated as subsequent placement, and most at last one group of image with
One page caricature distribution form is shown.When photograph album comprising amount of images than it is larger when, the present invention will be laid out with multipage caricature
Form carries out image shows.
By using the enumeration methodology of step 6, the present invention can obtain a series of character strings, and each character string will be right
Answer the original template of a caricature layout.For one group of data comprising 8 input pictures, it can be given birth to according to above-mentioned enumeration definition
Into 120 or so original templates.Fig. 5 is given for 8 input pictures, using the present invention enumerate obtained by character string and
Corresponding template situation.S in wherein Fig. 51, Si, SjRepresent to enumerate the L in obtained character string, character stringNRepresent template
Level situation (N maximums are 4, i.e., can at most be divided into four layers) residing for middle each unit lattice, LNContent correspondence in unquote is each
Cell enumerated strings situation, represents the particular location situation in each layer of each unit lattice in template.Shown by Fig. 5
T1, Ti, TjAs character string S1, Si, SjCorresponding actual template situation.All character strings as obtained by will enumerate are turned
Change, the present invention just can obtain corresponding dynamic template storehouse.
After dynamic template storehouse is obtained, the present invention wishes that the suitable template of selection is pieced together representative image, completes figure
As displaying.Representative image is carried out equal proportion scaling by the present invention according to page-size first, and it is then placed into mould in order
In plate each unit lattice, and require in picture centre and corresponding cell center alignment, such as Fig. 6 that b is shown after center alignment and put
Put situation.By observation as can be seen that in above process, due to the difference of each representative image style of shooting, making after its scaling
Size also will be different, so covering, gap and height will occur not etc. in each unit compartment in placement process
Phenomenon.In order to select template the most suitable in ATL, next, maximum principle will be presented in the present invention according to information content,
The preceding top-k template matched the most is searched in dynamic template storehouse by defining penalty, for users to use.
Introduce before penalty, it is as follows that the present invention defines each element symbol:For one group of image currently to be pieced together, if its
The original template storehouse tried to achieve isFor one of template Tm, its layer included is expressed as Li(i ∈ (1 ... 4)), Li
In cell be expressed as Pi,j(j ∈ (1 ... 3)), its implication is that the cell belongs to template TmI-th layer in j-th
Put, the rectangle that the representative image after scaling is put into after corresponding cell is expressed as Ii,j.Next dependent office will be considered
Portion and global restriction relation define appropriate factor of influence.
1):Layer LiThe coverage of interior each unit lattice.When being laid out, the present invention wishes to retain image as far as possible
All information, it is desirable that the area coverage of adjacent cells lattice is few as much as possible in layer, therefore definition layer LiInterior adjacent cells lattice
Level of coverage is after assignment:
Wherein Eo(Li) for i-th layer required by current template of area coverage, the cells that is included of the J for the template at i layers
Number, Area (Ii,j) it is the image area for being currently put into i j-th of cell of layer, Area (Ii,j+1) it is currently to be put into i layers of jth+1
The image area of individual cell.
2):Layer LiThe gap degree of interior each unit lattice.Equally when being laid out, if with each unit lattice after interlayer assignment
Gap is excessive, and the present invention need to readjust the ratio of representative image, so the gap degree with interlayer each unit lattice should be as far as possible
It is small, will not just destroy the integrality of picture.Therefore definition layer LiGap degree is after interior adjacent cells lattice assignment:
Wherein Eg(Li) for i-th layer required by current template of gap degree, J is the template in i-th layer of cell included
Number, Area (Li) it is i-th layer of area, Area (Ii,j) it is the image area for being currently put into i j-th of cell of layer, Area
(Ii,j+1) it is the image area for being currently put into+1 cell of i-th layer of jth.∩ operates to hand over, and represents that Two Areas area intersects
Situation, ∪ is and operated, and represents Two Areas area combination situation.
3):Layer LiThe degree of balance of interior each unit lattice.For being placed in cell adjacent in layer after image, such as two units
Its height difference is larger after lattice filling content, then destroys the harmony in layer, will cause every layer of disequilibrium, therefore the present invention
Height should try one's best and approach after each unit lattice filling content in desired layer, could so make it that layout seems more balance coordination.It is fixed
Adopted layer LiThe degree of balance of interior adjacent cells lattice is:
Wherein Eh(Li) for i-th layer required by current template of the degree of balance, the cells that is included of the J for the template at i-th layer
Number, Height (Ii,j) it is the picture altitude for being currently put into i-th layer of j-th of cell, Height (Ii,j+1) it is put into the to be current
The picture altitude of+1 cell of i layers of jth.
4) coverage of interlayer.Three constraintss only considered the local influence of each unit compartment in layer above, and grand
See that the whole structure of picture also relies on relation between layers in sight.For adjacent two layers up and down, if representative area
Being put into spilling border after cell causes area coverage between layers to destroy original image letter if excessive, not only
Breath, and have impact on the attractive in appearance of layout entirety.The covering that the present invention also is intended between layers is as far as possible small.Therefore definition layer LiWith
Layer Li+1Coverage be:
Wherein D (Li,Li+1) represent the coverage of i-th layer and i+1 layer, Area (Ii,j) it is put into j-th of list of i layers to be current
The image area of first lattice, Area (Ii+1,q) it is the image area for being currently put into i+1 q-th of cell of layer, Q is the list of i+1 layer
The quantity of first lattice;∩ operates to hand over, and represents that Two Areas area intersects situation.
Based on above-mentioned constraint definition, the present invention is for current template TmCorresponding penalty may be defined as following shape
Formula:
Wherein, Penalty (Tm) it is template TmPenalty, NL template TmThe maximum number of plies, LiRepresent i-th layer, Eo
(Li) it is i-th layer of area coverage, Eg(Li) it is i-th layer of gap degree, Eh(Li) it is i-th layer of the degree of balance, N is that Min-Max returns
One changes the factor;AREA is that the area of whole page caricature is used for normalization to area coverage between layers, α, β be local influence because
The weight of son and the global impact factor, α=0.4, β=0.6 is respectively set to by testing the present invention.
After top-k template is obtained, the present invention can be by the representative image after sequence again according to each unit lattice size
Scaled accordingly, obtain corresponding layout result, but because each cell of these templates is rectangle, to keep former
The basic proportionate relationship of image, it is difficult to accomplish completely to be filled by cell size, can still have covering, seam between image
The phenomenons such as gap, thus can cause layout result pattern lack change and unsightly the problems such as exist.In order to obtain more horn of plenty
Layout result, each unit lattice during the present invention is laid out to caricature carry out corresponding optimization deformation, retain original image area to maximize
Domain is principle, and carrying out cutting to representative image pieces together, to obtain the final layout result of pattern lively and changeable.
Because each unit lattice can respectively be split quadrangle obtained by straight line intersection by caricature cloth intra-office and constitute in template, a in such as Fig. 6
It is shown, therefore to adjustment each unit trellis shape, the present invention, which only needs respectively to split straight line to caricature cloth intra-office, to be adjusted accordingly i.e.
Can.The present invention is joined to retain the former region of representative image after scaling as far as possible as target by the segmentation straight line for adjusting Component units lattice
The optimization deformation (being the straight line that the present invention need to be adjusted such as the straight line identified in a in Fig. 6) for counting to complete each unit lattice.This
Invention setting final optimization pass target is as follows:
Wherein, TmFor current template;NL template TmThe maximum number of plies, NL ∈ (1...4);J ∈ (1...3) are every layer of institute
Comprising cell number;Pi,jRepresent that the cell is in template TmI-th layer in j-th of position;Ii,jFor by scaling
Representative image afterwards is put into the rectangle after i-th layer of j-th of cell;Ratio (x) (works as cell for the length-width ratio of corresponding region
When deforming upon, the value is the length-width ratio of minimum extraneous rectangle).The target of the present invention is minimum energy function Y (Tm)。
The present invention is using simple such as document 17:KENNEDY J.,EBERHART R.:Particle swarm
optimization.In Neural Networks,1995.Proceedings.,IEEE International
Conference on(1995),vol.4,pp.1942–1948vol.4.:The particle swarm optimization algorithm introduced is to current template
Object function optimize, the parameter corresponding to every segmentation straight line is encapsulated in template by particle, and by it each
Change in individual dimension and carry out optimization object function.After the object function reaches minimum or iteration certain number of times, the present invention will prolong shape
Straight line after change is cut to the image being put into, to obtain final layout result.As shown in c in Fig. 6, red frame is that image is final
Cutting situation, green frame is original template situation, it can be seen that each cell there occurs certain deformation, but to being placed
It is not very big that image majority, which cuts scope, can be good at retaining the most information of artwork.
Further, since in true caricature layout, can exist between layers, between cell and cell it is certain between
Away from making picture seem relatively sharp.Therefore, the present invention carries out simple movement to each straight line after deformation to produce correspondence effect
Really, the experimental result given by the present invention, interlamellar spacing is set to 15 pixels are wide, and it is wide that the spacing of cell is set to 8 pixels, such as schemes
In 6 shown in d, there is certain gap in each unit compartment, result is seemed more aesthetically pleasing.
The present invention relates to a kind of class caricature laying out images management method based on scene classification, how mainly solve to phase
Image carries out rational scene classification and how clearly two key problems of displaying is carried out to sorted image in volume.For
Scene classification problem, the present invention uses for reference the model tissue thought based on quadtrees (quartet analysis) qualitative analysis, right
Multiple global image characteristic use quaternary tree constructions that scene classification can be carried out have carried out quantitative analysis, and lead on this basis
Cross structure classification tree and realize the Rational Classification based on scene;For image shows problem after classification, the present invention is carried from classification tree
Take representative image and it is sorted, it is presented using with hierarchical sequence information, succinct caricature layout type.Its
Middle horn of plenty layout pattern, the present invention is also realized based on the dynamic template storehouse generation side enumerated by the design of caricature placement rule
Method, and the selection of template and piecing together automatically for image are completed based on the maximum principle of information content presentation to be shown image.This
The method of invention can be realized to the rationalization of image in photograph album and the target clearly shown.
Claims (8)
1. a kind of class caricature laying out images management method based on scene classification, it is characterised in that comprise the following steps:
Step one, ask for treating the characteristics of image of each image in organization chart picture set;
Step 2, quadtrees is built using the numerical distance between said features;
Step 3, passes through tried to achieve quadtrees and builds stratification classification tree;
Step 4, representative image is asked for using the classification degree between image in classification tree;
Step 5, is ranked up according to classification degree between representative image to representative image;
Step 6, is pieced together representative image quantity and the caricature rule of presence according to selected, is built using character string enumeration methodology
Dynamic template storehouse;
Step 7, matching template is searched for according to the maximum principle of information content that image after sequence is presented in dynamic template storehouse, and
Layout is optimized, final splicing result is obtained.
2. a kind of class caricature laying out images management method based on scene classification according to claim 1, it is characterised in that
Striked characteristics of image includes color, three features of shape and scene in step one.
3. a kind of class caricature laying out images management method based on scene classification according to claim 2, it is characterised in that
A constructed quadtrees includes two pairs of images in step 2, wherein every two images of centering are characterized in similar, and two
To being characterized in dissimilar between image.
4. a kind of class caricature laying out images management method based on scene classification according to claim 3, it is characterised in that
Constructed classification tree is a unrooted tree in step 3, and each leaf node of tree represents an image, non-leaf nodes generation
Table scene classification.
5. a kind of class caricature laying out images management method based on scene classification according to claim 4, it is characterised in that
Classification degree refers in classification tree an image to other images each shortest path to be undergone in step 4.
6. a kind of class caricature laying out images management method based on scene classification according to claim 5, it is characterised in that
The method generation dynamic template storehouse enumerated in step 6 using character string, is comprised the following steps:Layout order be set to from a left side to
It is right that the level of caricature where each cell is enumerated, the position of cell in every layer is enumerated from top to bottom, lead to
Cross two kinds and enumerate the corresponding caricature ATL of determination.
7. a kind of class caricature laying out images management method based on scene classification according to claim 6, it is characterised in that
Matching template is searched in dynamic caricature ATL according to the maximum principle of information content that image after sequence is presented in step 7, and
Layout is optimized, the preceding k template matched the most is searched in dynamic caricature ATL by defining penalty, including with
Lower step:Sequence representative image after scaling is put into each unit lattice in center alignment mode, adjacent unit compartment is produced
Raw gap, covering and difference in height is defined as local restriction relation;By between adjacent layer due to representative area overflow cell
And a part has been covered in next layer, the global restriction relation between adjacent layer is defined;According to part and global restriction relation,
By each template TmDefine penalty Penalty (Tm) be:
<mrow>
<mi>P</mi>
<mi>e</mi>
<mi>n</mi>
<mi>a</mi>
<mi>l</mi>
<mi>t</mi>
<mi>y</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>T</mi>
<mi>m</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>&alpha;</mi>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mi>L</mi>
</mrow>
</munderover>
<mrow>
<mo>(</mo>
<mi>N</mi>
<mo>(</mo>
<mrow>
<msub>
<mi>E</mi>
<mi>o</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
<mo>+</mo>
<mi>N</mi>
<mo>(</mo>
<mrow>
<msub>
<mi>E</mi>
<mi>g</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
<mo>+</mo>
<mi>N</mi>
<mo>(</mo>
<mrow>
<msub>
<mi>E</mi>
<mi>h</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&beta;</mi>
<mfrac>
<mrow>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mi>L</mi>
</mrow>
</munderover>
<mi>D</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>L</mi>
<mrow>
<mi>i</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>A</mi>
<mi>R</mi>
<mi>E</mi>
<mi>A</mi>
</mrow>
</mfrac>
<mo>,</mo>
</mrow>
Wherein:
<mrow>
<msub>
<mi>E</mi>
<mi>g</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>A</mi>
<mi>r</mi>
<mi>e</mi>
<mi>a</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>-</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>J</mi>
</munderover>
<mrow>
<mo>(</mo>
<mi>A</mi>
<mi>r</mi>
<mi>e</mi>
<mi>a</mi>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
<mo>&cap;</mo>
<mo>(</mo>
<mrow>
<mi>A</mi>
<mi>r</mi>
<mi>e</mi>
<mi>a</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>&cup;</mo>
<mi>A</mi>
<mi>r</mi>
<mi>e</mi>
<mi>a</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
1
<mrow>
<msub>
<mi>E</mi>
<mi>h</mi>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>J</mi>
</munderover>
<mo>|</mo>
<mi>H</mi>
<mi>e</mi>
<mi>i</mi>
<mi>g</mi>
<mi>h</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>&cap;</mo>
<mi>H</mi>
<mi>e</mi>
<mi>i</mi>
<mi>g</mi>
<mi>h</mi>
<mi>t</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>|</mo>
<mo>,</mo>
</mrow>
<mrow>
<mi>D</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>L</mi>
<mi>i</mi>
</msub>
<mo>,</mo>
<msub>
<mi>L</mi>
<mrow>
<mi>i</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>A</mi>
<mi>r</mi>
<mi>e</mi>
<mi>a</mi>
<mrow>
<mo>(</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>J</mi>
</munderover>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>&cap;</mo>
<mi>A</mi>
<mi>r</mi>
<mi>e</mi>
<mi>a</mi>
<mrow>
<mo>(</mo>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>q</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>Q</mi>
</munderover>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mo>+</mo>
<mn>1</mn>
<mo>,</mo>
<mi>q</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Penalty(Tm) represent template TmPenalty, α, β is respectively the weight of local restriction and global restriction, and N is min-
Max normalization factors, AREA is the size of caricature template, LiRepresent template TmI-th layer, NL represents template TmThe number of plies, Eo
(Li) it is i-th layer of area coverage, Eg(Li) it is i-th layer of gap degree, Eh(Li) it is i-th layer of the degree of balance;Ii,jExpression is put into
The presentation graphics of i j-th of cell of layer, Ii,j+1Expression is put into the presentation graphics of+1 cell of i-th layer of jth;Area
(Ii,j) it is the presentation graphics area for being put into i j-th of cell of layer, Area (Ii,j+1) it is to be put into+1 cell of i layers of jth
The area of presentation graphics;The number for the cell that J is included for i-th layer of the template, Area (Li) it is i-th layer of area;
Height(Ii,j) it is the picture altitude for being currently put into i-th layer of j-th of cell, Height (Ii,j+1) for it is current be put into i layers of jth+
The picture altitude of 1 cell;D(Li,Li+1) represent the coverage of i-th layer and i+1 layer, Area (Ii+1,q) it is currently to be put into i
The image area of+1 layer of q-th of cell, Q represents the cell quantity of i+1 layer;∩ operates to hand over, and represents two area surfaces
The intersecting situation of product;∪ is and operated, and represents two region area combination situations.
8. a kind of class caricature laying out images management method based on scene classification according to claim 7, it is characterised in that step
It is k template in rapid seven, retains representative area as object definition fitness function to maximize, and takes optimization method to suitable
Response function is optimized so as to generate the caricature layout that formwork style is abundant, is comprised the following steps:Select after k template,
It is for each template definition optimization aim in k template:
<mrow>
<mi>Y</mi>
<mrow>
<mo>(</mo>
<msub>
<mi>T</mi>
<mi>m</mi>
</msub>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>m</mi>
<mi>i</mi>
<mi>n</mi>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>i</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mrow>
<mi>N</mi>
<mi>L</mi>
</mrow>
</munderover>
<munderover>
<mo>&Sigma;</mo>
<mrow>
<mi>j</mi>
<mo>=</mo>
<mn>1</mn>
</mrow>
<mi>J</mi>
</munderover>
<mrow>
<mo>(</mo>
<mi>R</mi>
<mi>a</mi>
<mi>t</mi>
<mi>i</mi>
<mi>o</mi>
<mo>(</mo>
<msub>
<mi>P</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
<mo>-</mo>
<mi>R</mi>
<mi>a</mi>
<mi>t</mi>
<mi>i</mi>
<mi>o</mi>
<mo>(</mo>
<msub>
<mi>I</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mo>,</mo>
</mrow>
Wherein, TmFor current template;NL is template TmThe maximum number of plies, NL ∈ (1...4);J ∈ (1...3) are included by every layer
Cell number;Pi,jRepresent that the cell is in template TmI-th layer in j-th of position;Ii,jFor after scaling
Representative image is put into the rectangle after i-th layer of j-th of cell;Ratio (x) is the length-width ratio of corresponding region, is calculated using population
Method optimization energy function Y (Tm), final layout result is recommended into user.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410546275.5A CN104268580B (en) | 2014-10-15 | 2014-10-15 | A kind of class caricature laying out images management method based on scene classification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410546275.5A CN104268580B (en) | 2014-10-15 | 2014-10-15 | A kind of class caricature laying out images management method based on scene classification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104268580A CN104268580A (en) | 2015-01-07 |
CN104268580B true CN104268580B (en) | 2017-09-05 |
Family
ID=52160100
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410546275.5A Expired - Fee Related CN104268580B (en) | 2014-10-15 | 2014-10-15 | A kind of class caricature laying out images management method based on scene classification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104268580B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105511725A (en) * | 2015-12-09 | 2016-04-20 | 网易(杭州)网络有限公司 | Method and device for displaying controls in interface |
CN106056651A (en) * | 2016-06-03 | 2016-10-26 | 北京金山安全软件有限公司 | Picture layout management method and device and electronic equipment |
CN106709171B (en) * | 2016-12-13 | 2019-05-03 | 南京大学 | A kind of decalcomania generation method based on repeat pattern discovery |
CN108268838B (en) * | 2018-01-02 | 2020-12-29 | 中国科学院福建物质结构研究所 | Facial expression recognition method and facial expression recognition system |
CN112749538B (en) * | 2021-01-21 | 2022-08-12 | 厦门神州鹰软件科技有限公司 | Intelligent typesetting method and terminal |
CN116109893B (en) * | 2023-04-11 | 2023-09-15 | 宁波长壁流体动力科技有限公司 | Mine scene image classification method, system and storage medium |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101373517A (en) * | 2007-08-22 | 2009-02-25 | 北京万集科技有限责任公司 | Method and system for recognizing license plate |
CN101620742A (en) * | 2009-08-12 | 2010-01-06 | 国网电力科学研究院 | Real-time data processing method for three-dimensional rendering |
-
2014
- 2014-10-15 CN CN201410546275.5A patent/CN104268580B/en not_active Expired - Fee Related
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101373517A (en) * | 2007-08-22 | 2009-02-25 | 北京万集科技有限责任公司 | Method and system for recognizing license plate |
CN101620742A (en) * | 2009-08-12 | 2010-01-06 | 国网电力科学研究院 | Real-time data processing method for three-dimensional rendering |
Non-Patent Citations (1)
Title |
---|
"Image classification for content-based indexing";Vailaya A等;《IEEE Transactions on Image Proeessing》;20101031;117-129 * |
Also Published As
Publication number | Publication date |
---|---|
CN104268580A (en) | 2015-01-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104268580B (en) | A kind of class caricature laying out images management method based on scene classification | |
Zheng et al. | Content-aware generative modeling of graphic design layouts | |
Hu et al. | Recognition of pornographic web pages by classifying texts and images | |
CN109947987B (en) | Cross collaborative filtering recommendation method | |
Lim et al. | Context by region ancestry | |
CN105354593B (en) | A kind of threedimensional model sorting technique based on NMF | |
CN108009286A (en) | A kind of Sketch Searching method based on deep learning | |
CN108875076A (en) | A kind of quick trademark image retrieval method based on Attention mechanism and convolutional neural networks | |
Dewi et al. | Taiwan stop sign recognition with customize anchor | |
CN113032613B (en) | Three-dimensional model retrieval method based on interactive attention convolution neural network | |
Iqbal et al. | Feature integration, multi-image queries and relevance feedback in image retrieval | |
CN107944485A (en) | The commending system and method, personalized recommendation system found based on cluster group | |
CN107833224A (en) | A kind of image partition method based on multi-level region synthesis | |
Wang et al. | Remote-sensing image retrieval by combining image visual and semantic features | |
CN106778834A (en) | A kind of AP based on distance measure study clusters image labeling method | |
CN110737805A (en) | Method and device for processing graph model data and terminal equipment | |
CN114741519A (en) | Paper correlation analysis method based on graph convolution neural network and knowledge base | |
CN110349159A (en) | 3D shape dividing method and system based on the distribution of weight energy self-adaptation | |
CN102902976A (en) | Image scene classification method based on target and space relationship characteristics | |
Nie et al. | SHREC’18 track: 2D scene sketch-based 3D scene retrieval | |
CN109582783A (en) | Hot topic detection method and device | |
Qian et al. | LAST: Location-appearance-semantic-temporal clustering based POI summarization | |
Guo et al. | Multigranular event recognition of personal photo albums | |
CN111652309A (en) | Visual word and phrase co-driven bag-of-words model picture classification method | |
Chen et al. | HyNet: A novel hybrid deep learning approach for efficient interior design texture retrieval |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170905 Termination date: 20181015 |
|
CF01 | Termination of patent right due to non-payment of annual fee |