CN106202352B - The method of indoor furniture style and colour match design based on Bayesian network - Google Patents
The method of indoor furniture style and colour match design based on Bayesian network Download PDFInfo
- Publication number
- CN106202352B CN106202352B CN201610525746.3A CN201610525746A CN106202352B CN 106202352 B CN106202352 B CN 106202352B CN 201610525746 A CN201610525746 A CN 201610525746A CN 106202352 B CN106202352 B CN 106202352B
- Authority
- CN
- China
- Prior art keywords
- color
- furniture
- style
- scene
- bayesian network
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/583—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content
- G06F16/5838—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using metadata automatically derived from the content using colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
- G06F18/24155—Bayesian classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Library & Information Science (AREA)
- Probability & Statistics with Applications (AREA)
- Databases & Information Systems (AREA)
- Image Analysis (AREA)
- Processing Or Creating Images (AREA)
Abstract
The method of the indoor furniture style and colour match design that the invention discloses a kind of based on Bayesian network the, comprising steps of 1) collection of room style and color design scheme；2) design scheme of downloading is labeled；3) color of every type objects is clustered；4) training Bayesian network；5) color recommendation is carried out for 3D indoor scene；6) color diversification adjusts；7) on the color transfer to texture that design obtains, and texture is attached on 3D model of place.Relationship of the present invention using Bayesian network from outstanding interior design plan (picture) between coding decorated style and furniture colour match, can be used for the design of the furniture colour match in indoor scene.Requirement according to user to decorated style, system of the invention can recommend the furniture colour match for meeting the indoor scene of the style out, conveniently effect displaying is provided for the house ornamentation market in China city, 3D game animation and virtual reality scenario, there is actual promotional value.
Description
Technical field
The present invention relates to the technical fields that automation collocation design is carried out to indoor furniture color, refer in particular to a kind of base
In the method for the indoor furniture style of Bayesian network and colour match design.
Background technique
Indoors in design, in addition to the layout of consideration furniture, color design is also a critically important factor.Psychology is ground
Study carefully the mood for pointing out that color can influence people, certain emotion of people can be excited.For example, warm colour normally convey warmness and actively
Upward mood；And cool colour then makes us feeling refrigerant.In the influence to space, cool colour can extend the space of cubicle away from
From；Warm colour makes a big room seem dynamic.However, to the object of 3 dimension spaces carry out coloring be one very
There is the problem of challenge.Factor in need of consideration includes the influence to space of the hobby of inhabitation people, the function in room, color, color
Collocation and color itself etc..
There are many available empirical rules for traditional indoor design.For example, to the coloring in room from the color of entire scene
Impression starts, or starts (contraction and inflated sense of cold-warm color) using the defect that color makes up space, or according to certain
Sequence: the background colours such as metope and floor, primary role, complementary colors and intersperse color this four roles and successively colour, and colouring
The color matching relationship between them is considered in journey.For a professional designer, by manually determining each family in room
The color of tool is an extremely difficult and cumbersome task.It is in room especially for no trained ordinary user
Furniture selects a color just more difficult.So auto color recommender system is for helping non-professional ordinary user to set
The meter interior space plays very important meaning and effect.
Currently, there are also the work of similar colour match.But, application scenarios are different.Have plenty of to 3D object
Different components coloured [A.Jain, T.Thormahlen, T.Ritschel, H.-P.Seidel, Material memex:
automatic material suggestions for 3d objects,ACM Transactions on Graphics
(TOG) 31 (6) (2012) 143.], have plenty of to the pattern of 2D image carry out colour match design [S.Lin, D.Ritchie,
M.Fisher,P.Hanrahan,Probabilistic color-by-numbers:Suggesting pattern
colorizations using factor graphs,ACM Transactions on Graphics(TOG)32(4)
(2013) 37.], also some for 3D indoor scene provide material recommend [K.Chen, K.Xu, Y.Yu, T.-Y.Wang, S.-M.Hu,
Magic decorator:automatic material suggestion for indoor digital scenes,ACM
Transactions on Graphics (TOG) 34 (6) (2015) 232.], but they are all without reference to decoration style.And it is real
On border, when house is fitted up and selects furniture, house owner can consider oneself to select difference to the hobby of color and style
Stylistic category.Therefore, present system provides the selections of different-style, to recommend the colour match side of furniture for indoor scene
Case.
Summary of the invention
It is an object of the invention to overcome the design of the color of current indoor scene to depend on hand-designed, others are certainly
Dynamic mode otherwise applied to complex object components coloring or be merely given as the general scene of no decoration style design
Material the problem of recommending, a kind of method for providing the indoor furniture style based on Bayesian network and colour match design,
The automatic collocation of furniture color can be realized according to home decoration style to the good with colour cast of color.
To achieve the above object, technical solution provided by the present invention are as follows: the indoor furniture style based on Bayesian network
With the method for colour match design, comprising the following steps:
1) collection of room style and color design scheme
Color design scheme (the figure of different-style and different room types is downloaded from existing interior decoration site for service
Piece), as the database for machine learning；
2) design scheme of downloading is labeled
Scene one by one manually extracts the color of each furniture, the color including metope, floor；
3) color of every class furniture is clustered
The classification for extracting furniture in scene carries out the cluster of two levels to the color of every class furniture, first according to different
Style once clusters each type of furniture；Then, according to every kind of furniture in the color cluster knot for summarizing all styles
Under fruit, the color of every kind of furniture is clustered once again；
4) training Bayesian network
By marking obtained color and its cluster as a result, pre-processed to obtain the record of each scene, this is utilized
A little data training Bayesian networks, obtain Bayesian network model and its conditional probability distribution；
5) color recommendation is carried out for 3D indoor scene
Virtual 3D scene is established, and utilizes in step 4) and obtains the face in conditional probability distribution and step 3) about style
Color cluster result establishes the objective function for meeting colour match design, is in household scene using Simulated Anneal Algorithm Optimize
Each furniture recommends color rendering intent out；
6) color diversification adjusts
The color category of each furniture obtained in step 5) is limited, in order to diversified and do not lose original style
Design, the local coordinate (principal component corresponding to RGB) of each color category is obtained using PCA, right under this local coordinate
Color disturbs, and generates other colors abundant；
7) on the color transfer to texture that design obtains, and texture is attached on 3D model of place
The color of each furniture obtained in step 6) is moved to first on the optional texture of user, then by color
In texture mapping to single 3D model after migration.
Each step is described in detail below.
In step 2), special marking program is needed, " frame choosing " is utilized to each furniture in the picture of downloading first
Its color is extracted in operation.The size of choice box, by mark personnel depending on the size of furniture in picture." frame choosing " the result is that
Certain partial pixel region of furniture, calculate region colored pixels average value can obtain the mark furniture primary color.
When preservation, as unit of each scene, it is considered as a record.In a record, scene number, scene type, finishing are recorded
The color that style (being denoted as S) and each furniture mark.
In step 3), in order to make the color category obtained after furniture cluster closer to every class style, to the color of furniture
Carry out the cluster of two levels.Firstly, according to the classification of style, i.e., under certain style, using K-means method to every class man
The color of tool clusters once respectively, has obtained at style S, and the color cluster of kth class furniture is as a result, be denoted asK=
1,...,N；Then, according to every kind of furniture in the case where summarizing the color cluster result of all styles, the color of every kind of furniture is gathered again
Class is primary；
The selection of cluster centre: being changed by section of the setting cluster centre number from 2 to 9, observe all sample points from
Whether the curvilinear trend of the average weight distance of its corresponding cluster centre is stable to determine.For example, if cluster number is 5
When, the curve with a distance from average weight of all sample points from its corresponding cluster centre is a turning point；Later, all sample points
The average weight distance value of cluster centre corresponding from its tends towards stability, then cluster number just selects 5.
In step 4), after step 3) obtains the color of each furniture, need to handle the scene note saved originally again
Record, i.e., obtain certain corresponding color after the primitive furniture color marked to each furniture being replaced with color cluster
(cluster centre).The result of replacement is denoted as: D={ di, i=1,2 ..., M }, whereinAnd then, will
Obtained scene record is established and the consistent network structure of scene record data and calculating for training TAN Bayesian network
Condition probability distribution table out.
Note B=< G, Θ > is TAN Bayesian network, and G indicates a directed acyclic graph, Θ expression parameter collection.Wherein, G
({ A is denoted as comprising 15 stochastic variable nodes0,A1,...,Ak) and node between directed edge collection.Particularly, A0It is expressed as classification
Node, AkIndicate the furniture of k-th of type；Θ is embodied as depending on the probability on node, i.e., for variables AkPossibility take
Value ak∈Ck(CkFor the set for the color category that kth class furniture can choose) and its father node possibility value
HaveWherein,Represent AkFather node set.
Create TAN Bayesian network: firstly, establishing weight limit spanning tree.Weight in spanning tree between two nodes is
Conditional mutual information amount, is calculated as follows mode:
Wherein,Expression experience is distributed, and in realization, statistic frequency is as its probability.
Then, it selects one of node for root, spanning tree is converted into directed tree, other nodes are directed toward from root in direction.
Then, category node is added, and increases from category node to the directed edge of each attribute node.This just establishes TAN shellfish
This network of leaf.Finally, being calculated whole according to the data D that the TAN network of foundation and pretreatment obtain using experienced probability distribution
The probability distribution of a Bayesian network.
In step 5), user can be allowed to specify some other color of (a little) furniture.It will be by the family of user specified color
Have category set, is denoted as K；The furniture classification of non-fundamental color is denoted as U.Without loss of generality, remember ak∈CkCertain face specified by algorithm
Color, wherein k ∈ U.Need to find mapping: A for all k ∈ Uk→ak.Finally, the color to furniture each in scene
Selection is expressed as the problem of minimizing objective function:
E (Φ)=w1Eindiv(Φ)+w2Ejoint(Φ)+w3Eclose(Φ) (1)
In formula, Φ={ Ak,k∈U}。
In catalogue scalar functions, 3 sub-goals are defined:
1. maximizing the probability that single furniture color occurs.In the color condition that given user is the selection of some (a little) furniture
Under, select the conditional probability of certain color to be defined as P (A for other single furniturek|{Aj=aj,j∈K}).The target attempts to be every
The probability that the color of a furniture selection occurs in training data is as big as possible, that is, minimizes following objective function:
2. maximizing the joint probability of colour match between furniture.Color is selected for each furniture, so that the face of entire scene
Color collocation looks like coordination.I.e. in training data, most commonly-encountered furniture colour match selects joint probability highest
Color combination:
Ejoint(Φ)=1-P ({ Ak,k∈U}|{Al=al,l∈K}) (3)
3. minimizing at a distance from decorated style (color).In decoration practice indoors, decorated style is generally referred to by user
It is fixed, i.e. A0∈C0It is known.Every kind of decorated style has several colors of its preference.In order to make the color for assigning each furniture whole
As consistent with decorated style as possible on body, the color of the every kind of furniture that needs restraint and the primary color of selected decorated style are got over
It is close better, i.e.,
Compared with prior art, the present invention have the following advantages that with the utility model has the advantages that
1, the method for the present invention introduces concept of the indoor design field about decorated style.In the data set of study, press
Classify according to decorated style.The Bayesian network model trained can encode interior between decorated style and each furniture color
It is contacting.In this way, the solid colour that the furniture color for coming out system recommendation is preferred with decorated style, meets designers
Design effect.
2, the method for the present invention utilizes Bayesian network from the collocation learnt between furniture color in outstanding interior design plan
Relationship meets beautiful design so that the furniture colour match in the indoor scene that system recommendation obtains seems to coordinate, is good-looking
Demand.
3, fixed particularly suitable for currently a popular product on the texture that the method for the present invention can like color transfer to user
Demand processed.This system recommends the color of indoor coordination out, and user can choose the texture oneself liked, by the mass-tone decantation of color
It moves on on the texture of selection, meets the individual demand of user.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention.
Fig. 2 is the example for the interior design plan (picture) that present invention collection obtains.
Fig. 3 is the interface that the present invention is labeled the design scheme (picture) of downloading.
Fig. 4 is the partial results figure that the furniture color that the present invention obtains mark is clustered.
Fig. 5 is that the result figure that one embodiment of the present of invention obtains Bayesian network model training (only shows main section
Point and key side).
Fig. 6 is the comparison diagram of the color disturbance front and back of one embodiment of the present of invention.
Fig. 7 is comparison figure of the one embodiment of the present of invention before and after texture is added.
Fig. 8 is the effect picture for the Chinese style style that one embodiment of the present of invention is recommended.
Fig. 9 is the effect picture for the European style that one embodiment of the present of invention is recommended.
Figure 10 is the stylish effect picture that one embodiment of the present of invention is recommended.
Figure 11 is the effect picture for the neo-pragmatism style that one embodiment of the present of invention is recommended.
Figure 12 is the effect picture for the pastoralism that one embodiment of the present of invention is recommended.
Figure 13 is that the design scheme that recommender system of the invention generates obtains the system correctly classified for 5 SVM classifiers
Count result figure.
Phase of the Figure 14 for the scheme of user's evaluation present system recommendation and the scheme of former designer's design on dominant hue
Like the average mark of degree and the statistical results chart of 95% confidence interval.
Figure 15 is scheme, the scheme of designer's design of invitation, the side recommended at random that user recommends present system
The average mark of the visual quality evaluation of case and the statistical results chart of 95% confidence interval.
Figure 16 is the vision matter of scheme and the Magic Deocorator scheme recommended that user recommends present system
Measure the average mark of evaluation and the statistical results chart of 95% confidence interval.
Specific embodiment
The present invention is further explained in the light of specific embodiments.
As shown in Figure 1, what the indoor furniture style and colour match described in the present embodiment based on Bayesian network designed
Method, comprising the following steps:
1) collection of room style and color design scheme
Different-style and not chummery class are downloaded from interior decoration site for service " Shang Pinzhai matches " homepage of domestic profession
The color design scheme (picture) of type, as the database for machine learning.One design scheme example of downloading: interior is not
Picture and corresponding furniture type with visual angle, as shown in Figure 2.From these data of downloading designer's scheme, three are summarized
The feature of group data, such as following table.
2) the design scheme picture of downloading is labeled
Scene one by one extracts the color of each furniture, the color including metope, floor.As shown in figure 3, this is that we open
The software interface of the furniture color mark of hair.
3) color of every class furniture is clustered
The classification for extracting furniture in scene carries out the cluster of two levels to the color of every class furniture, first according to different
Style once clusters each type of furniture；Then, according to every kind of furniture in the color cluster knot for summarizing all styles
Under fruit, the color of every kind of furniture is clustered once again；Finally, the color cluster of obtained main furniture is as a result, as shown in Figure 4.
4) training Bayesian network
By marking obtained color and its cluster as a result, pre-processed to obtain the record of each scene, this is utilized
A little data training Bayesian networks, obtain Bayesian network model and its conditional probability distribution.The Bayes that training obtains
Network model schematic diagram, as shown in Figure 5.
5) color recommendation is carried out for 3D scene
Virtual 3D scene is established, and is obtained in conditional probability distribution and step 3) using training in step 4) for furniture
Color about style color cluster as a result, establish meet colour match design energy function as objective function.Work as user
After specifying style for scene and be a or two a furniture designated color, using Simulated Anneal Algorithm Optimize, so that system is household
Each furniture in scene recommends colour match scheme out, two picture of the left side of two example corresponding diagrams 6 therein, wherein upper left
The decorated style of scene belongs to Chinese style style, and the decorated style of lower-left belongs to neo-pragmatism style.
6) color diversification adjusts
The color category of each furniture obtained in step 5) is limited, in order to diversified and do not lose original style
Design, the local coordinate of each color category is obtained using PCA, disturbs under this local coordinate to color, generates other
Color abundant.Effect after disturbing to color, as shown in Figure 6, the left side are two obtained the scenes through directly optimizing, right
While being corresponding two scenes after disturbance, centre is the comparison that main furniture color disturbs its color block of front and back.
7) on the color transfer to texture that design obtains, and texture is attached on 3D model of place
Texture fit up indoors in effect it is extremely important.There is no the world of texture to seem untrue.So by step
6) color of each furniture obtained in is moved to first on the optional texture of user, then by the texture mapping to single 3D
On model.Sticked on 3D mock-up funiture before and after texture as a result, as shown in fig. 7, two scenes on the left side are to go up color
As a result, two scenes on the right are to stick the result of texture, hence it is evident that, two scenes on the right seem more there is the sense of reality.
Method in order to verify proposition of the invention, has done following experiment:
1) visual effect.5 model of place of corresponding 5 kinds of stylistic categories are established to be verified, final system is recommended
Effect after the result obtained is rendered, is shown in Fig. 8, Fig. 9, Figure 10, Figure 11, Figure 12, respectively correspond Chinese style style, European style,
Modernism style, neo-pragmatism style and pastoralism.
2) whether the color recommendation results that verifying system obtains meet certain style.In order to answer this problem, 5 are had trained
A SVM classifier.Each SVM classifier is responsible for identifying a kind of style.Firstly, the data set that former mark obtains is divided into two groups:
80% is used to test for training and 20%.For example, will indicate Chinese style style is positive sample, other styles are the number of negative sample
According to for training the classifier of an identification Chinese style style.The furniture color data for each scene that these marks are obtained, gathers
At 5 kinds of theme colors, using this 5 kinds of theme colors extract 334 dimensional features [P.O'Donovan, A.Agarwala,
A.Hertzmann,Color compatibility from large datasets,ACM Transactions on
Graphics (TOG) 30 (4) (2011) 63.], for training SVM classifier.With 20% data test trained classification
The reliability of device.Similarly, using the method for the present invention, for 20 scenes of every kind of Style Design.In each experiment, user
Style specified first and the color for assigning 1-2 different furniture, recommender system select mutual as other still uncoloured furniture
The mutually color of collocation.5 kinds of styles generate 100 scenes altogether.Each scene includes the result directly optimized and disturbs through color
Result.The two results extract 5 kinds of theme colors respectively, extract 334 dimensional features again using this 5 kinds of theme colors, are used for
Test trained 5 SVM classifiers.As shown in Figure 13, " C, E, M, N, P " respectively indicate Chinese style style, European to test result
Style, smartness, neo-pragmatism style, 5 classifiers corresponding to pastoralism, first, second and third histogram difference
It indicates 20% test data, directly optimize Chinese style classifier corresponding to obtained color data, the color data after disturbing
Accuracy rate.It can be seen from the figure that 20% test data classification accuracy rate all 83% or more, and by the present invention propose
The resulting data of method, Accurate classification rate has been more than 80%.That is, the recommendation results proposed are close to decorated style
Color expression, ironically, by disturbance can improve to the classification accuracy for directly optimizing obtained color data.
3) user investigation is studied.The result for evaluating the recommendation of a furniture color is extremely difficult.For this purpose, selection user investigation is ground
The mode studied carefully, which verifies the method for the present invention, can be comparable to Specialty Design teacher, and be better than the method for random colouration.
Firstly, to verify Style Design caused by color recommender system of the invention meets the style that user specifies.It invites
Please 30 people, take part in this user investigation.Questionnaire shares 20 problems, and each problem includes 3 pictures.Intermediate figure
Piece derives from tranining database, and the picture on both sides is then generated by method of the invention and random method.Both methods is being calculated
Input before method operation is equally, that is, user to be needed to specify identical 2-3 furniture color.The picture that both methods generates exists
Putting in order for both sides is deviation that is random, generating to avoid sequence of positions.Participant is invited to evaluate both sides picture and centre
Picture similar degree in shade of color.0 point indicates not at all as 5 points of expressions are as very much.Finally, the result warp of investigation
Statistics calculating, method of the invention and the design scheme similarity degree learnt are equally divided into 3.4, higher than random method
(1.9 points).For every class style scene, Figure 14 shows the confidence in 2 kinds of methods to the average mark of 5 scene statistics and 95%
Section.The result of T-test illustrates method of the invention far better than random method.
Result that second experiment designs the results of our methods, Specialty Design teacher for comparing user and random
The result of colouration method is evaluated.Similarly, 5 virtual scenes are established respectively, correspond to 5 kinds of room styles.For every
A virtual scene, we have invited Specialty Design teacher to do the design for meeting a certain style using professional software；Meanwhile
4 designs are also created according to different inputs with method of the invention；In addition, not considering style, also obtained with random method
The result of a colouration out.The 3D scene that all these three methods obtain is rendered under identical visual angle and illumination condition to be obtained
2D image.The images of corresponding 6 renderings of each virtual scene, wherein 1 from designer's design, 4 from of the invention
Method, another result by random colouration.In user investigation questionnaire, the image of a total of 30 renderings.In order to avoid by
In the deviation that sequence generates, this 6 images are arranged in every topic with random sequence.Then, 80 volunteers has been invited to participate in adjusting
It looks into, inquires the visual quality of their every image, them is asked to evaluate 0 (being difficult to see) point to 5 points (very good-looking).Finally, statistics
The result is shown in Figure 15 shown in: average mark and corresponding 95% confidence interval of three kinds of methods to 5 class scenes, and to all scenes
Grand average and corresponding 95% confidence interval.The score average mark of the result of designer's design is only than result of the invention
Have more 0.1.In addition, examined using z-test, our result score is better than random, but is no more than 24% (p-value=
0.028), though the result score of designer's design is better than us, but it is no more than 7% (p-value=0.031).
4) compared with newest method.Magic Decorator[K.Chen,K.Xu,Y.Yu,T.-Y Wang,S.-
M.Hu,“Magic decorator:automatic material suggestion for indoor digital
Scenes, " in ACM Transactions on Graphics, vol.34, no.6,2015, p.232.] can directly carry out
Material is recommended, including material properties and texture.Although method of the invention only does color recommendation, pass through color transfer to texture
Mode on picture provides material texture for user.In order to complete this experiment, the author of Magic Decorator is invited to come
Help is that 5 bedroom scenes respectively generate 2 results using Magic Decorator method.Equally, using method of the invention
2 results are also respectively generated according to identical input condition.Here input condition is the color of a or two a furniture specified in advance
As constraint condition.These obtained 3D scenes are rendered under identical visual angle and illumination condition, obtain 2D image.With 3)
In experiment way it is the same, invite same a group of people to participate in investigation, score to the visual quality of every image.Questionnaire
5 problems are shared, every problem 4 opens image, wherein two methods from Magic Decorator, another two from the present invention
Method.The average mark of 5 scenes of statistics and 95% confidence interval are as shown in Figure 16.It is tested using z-test, this hair
The scoring of bright method acquired results and the scoring of the result of Magic Decorator method are suitable (p-value=0.48).
The method of the present invention is by above the experiment proves that its feasibility, superiority.What although example of the invention was directed to
It is the scene in bedroom, but can extend to other scenes, such as parlor, kitchen, office etc., on condition that needing to collect it
The design scheme data of his scene.
In conclusion the present invention from the design scheme (picture) of designer by the method for machine learning, use pattra leaves
This network model encodes the inner link of household style and furniture colour match, and guidance furniture color recommender system generation meets certain
The scene of kind decorated style.The visual quality that the generated scene of the method for the present invention is demonstrated in such a way that user investigation is studied connects
It is bordering on the design effect of Specialty Design teacher, there is actual promotional value.
The examples of implementation of the above are only the preferred embodiments of the invention, and implementation model of the invention is not limited with this
It encloses, therefore all shapes according to the present invention, changes made by principle, should all be included within the scope of protection of the present invention.
Claims (2)
1. the method for indoor furniture style and colour match design based on Bayesian network, which is characterized in that including following step
It is rapid:
1) collection of room style and color design scheme
Color design scheme, that is, picture of different-style and different room types is downloaded from existing interior decoration site for service,
As the database for machine learning；
2) design scheme of downloading is labeled
Scene one by one manually extracts the color of each furniture, the color including metope, floor；Wherein, special mark journey is needed
Sequence: " frame choosing " operation is utilized to extract its color in each furniture in the picture of downloading first；The size of choice box, by marking people
Member is depending on the size of furniture in picture；" frame choosing " the result is that certain partial pixel region of furniture, calculates the face in region
Color pixel average value can obtain the primary color of the furniture of the mark；When preservation, as unit of each scene, it is considered as a note
Record；In a record, color that record scene number, scene type, decoration style and each furniture mark；
3) color of every class furniture is clustered
The classification for extracting furniture in scene carries out the cluster of two levels to the color of every class furniture, first according to different styles,
Each type of furniture is once clustered；Then, according to every kind of furniture in the case where summarizing the color cluster result of all styles,
The color of every kind of furniture is clustered once again；
In order to make the color category obtained after furniture cluster closer to every class style, the poly- of two levels is carried out to the color of furniture
Class: it firstly, according to the classification of style, i.e., under certain style, is clustered respectively using color of the K-means method to every class furniture
Once, it has obtained at style S, the color cluster of kth class furniture is as a result, be denoted asThen, according to every
Kind furniture clusters once the color of every kind of furniture in the case where summarizing the color cluster result of all styles again；
The selection of cluster centre: changed by section of the setting cluster centre number from 2 to 9, it is right from its to observe all sample points
Whether the curvilinear trend of the average weight distance for the cluster centre answered is stable to determine；
4) training Bayesian network
By marking obtained color and its cluster as a result, pre-processed to obtain the record of each scene, these numbers are utilized
According to training Bayesian network, Bayesian network model and its conditional probability distribution are obtained, specific as follows:
After step 3) obtains the color of each furniture, need to handle the scene record saved originally again, i.e., it will be to each family
The primitive furniture color that tool mark obtains obtains corresponding certain color i.e. cluster centre after replacing with color cluster, replaces
Result be denoted as: D={ di, i=1,2 ..., M }, whereinAnd then, obtained scene is recorded
For training TAN Bayesian network, it is therefore an objective to establish and record the consistent network structure of data with scene and to calculate condition general
The process for establishing TAN Bayesian network is described in detail below in rate distribution table:
Note B=< G, Θ > is TAN Bayesian network, and G indicates a directed acyclic graph, Θ expression parameter collection；Wherein, G includes
15 stochastic variable nodes are denoted as { A0,A1,...,AkAnd node between directed edge collection；A0It is expressed as category node, AkIndicate kth
The furniture of a type；Θ is embodied as depending on the probability on node, i.e., for variables AkPossibility value ak∈CkAnd his father
The possibility value of nodeHaveWherein,Represent AkFather node set；
Creation TAN Bayesian network: firstly, establishing weight limit spanning tree, the weight in spanning tree between two nodes is condition
Mutual information is calculated as follows mode:
Wherein,Expression experience is distributed, and in realization, statistic frequency is as its probability；
Then, it selects one of node for root, spanning tree is converted into directed tree, other nodes are directed toward from root in direction；Then,
Category node is added, and increases from category node to the directed edge of each attribute node；This just establishes TAN Bayesian network
Network；Finally, calculating entire pattra leaves using experienced probability distribution according to the data D that the TAN network of foundation and pretreatment obtain
The probability distribution of this network；
5) color recommendation is carried out for 3D indoor scene
Virtual 3D scene is established, and poly- about the color of style in conditional probability distribution and step 3) using obtaining in step 4)
Class is as a result, establish the objective function for meeting colour match design, using Simulated Anneal Algorithm Optimize, for each family in household scene
Tool recommends color rendering intent out；
6) color diversification adjusts
The color category of each furniture obtained in step 5) is limited, in order to diversified and do not lose setting for original style
Meter, the local coordinate of each color category is obtained using PCA, disturbs under this local coordinate to color, it is rich to generate other
Rich color；
7) on the color transfer to texture that design obtains, and texture is attached on 3D model of place
The color of each furniture obtained in step 6) is moved to first on the optional texture of user, then by color transfer
In texture mapping afterwards to single 3D model.
2. the method for the indoor furniture style according to claim 1 based on Bayesian network and colour match design,
It is characterized in that: in step 5), user can be allowed to specify some or the other color of some furnitures, it will be by user specified color
Furniture category set, is denoted as K；The furniture classification of non-fundamental color is denoted as U；Remember ak∈CkCertain color specified by algorithm, wherein k
∈U；Need to find mapping: A for all k ∈ Uk→ak；Finally, the color to furniture each in scene is selected expression
The problem of to minimize objective function:
E (Φ)=w1Eindiv(Φ)+w2Ejoint(Φ)+w3Eclose(Φ) (1)
In formula, Φ={ Ak,k∈U}；
In catalogue scalar functions, 3 sub-goals are defined:
1. maximizing the probability that single furniture color occurs: in the color condition that given user is some or the selection of some furniture
Under, select the conditional probability of certain color to be defined as P (A for other single furniturek|{Aj=aj,j∈K})；The target attempts to be every
The probability that the color of a furniture selection occurs in training data is as big as possible, that is, minimizes following objective function:
2. maximizing the joint probability of colour match between furniture: color is selected for each furniture, so that the color of entire scene is taken
With coordination is looked like, i.e., in training data, most commonly-encountered furniture colour match selects the highest face of joint probability
Colour cell is closed:
Ejoint(Φ)=1-P ({ Ak,k∈U}|{Al=al,l∈K}) (3)
3. minimizing at a distance from decorated style, that is, color: in decoration practice indoors, decorated style is usually specified by user, i.e.,
A0∈C0It is known；Every kind of decorated style has several colors of its preference；In order to keep the color for assigning each furniture most on the whole
May be consistent with decorated style, the color of the every kind of furniture that needs restraint and the primary color of selected decorated style are closer to more
It is good, i.e.,
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610525746.3A CN106202352B (en) | 2016-07-05 | 2016-07-05 | The method of indoor furniture style and colour match design based on Bayesian network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610525746.3A CN106202352B (en) | 2016-07-05 | 2016-07-05 | The method of indoor furniture style and colour match design based on Bayesian network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106202352A CN106202352A (en) | 2016-12-07 |
| CN106202352B true CN106202352B (en) | 2019-11-15 |
Family
ID=57465313
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610525746.3A Active CN106202352B (en) | 2016-07-05 | 2016-07-05 | The method of indoor furniture style and colour match design based on Bayesian network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106202352B (en) |
Families Citing this family (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI613736B (en) * | 2017-02-18 | 2018-02-01 | 盟立自動化股份有限公司 | Carrier structure and carrier substrate thereof |
| CN108694266A (en) * | 2017-04-07 | 2018-10-23 | 美宅科技（北京）有限公司 | intelligent decoration design method and system based on machine learning |
| US10824317B2 (en) * | 2017-06-14 | 2020-11-03 | Behr Process Corporation | Systems and methods for assisting with color selection |
| CN107562963B (en) * | 2017-10-12 | 2021-04-20 | 杭州群核信息技术有限公司 | Method and device for screening home decoration design rendering graph |
| CN107818341A (en) * | 2017-10-25 | 2018-03-20 | 天津大学 | A kind of color extraction method based on improvement K means algorithms |
| CN108846904A (en) * | 2018-06-28 | 2018-11-20 | 广州奥韵荟家居饰品发展有限公司 | Spatial form Intelligentized design method and system |
| CN110738715B (en) * | 2018-07-19 | 2021-07-09 | 北京大学 | Automatic migration method of dynamic text special effect based on sample |
| CN109522439B (en) * | 2018-10-17 | 2021-07-09 | 东软集团股份有限公司 | Label retrieval method, device, storage medium and equipment for design resources |
| JP2020087224A (en) * | 2018-11-29 | 2020-06-04 | 富士ゼロックス株式会社 | Information processing device and information processing program |
| CN109636535A (en) * | 2018-12-19 | 2019-04-16 | 杭州庭好数字科技有限公司 | Fit up set meal customization of individual character system and method |
| CN109947976B (en) * | 2019-01-22 | 2023-05-16 | 湖南城市学院 | System and method for analyzing Chinese traditional element embodiment in modern art design |
| CN110083430B (en) * | 2019-04-30 | 2022-03-29 | 成都映潮科技股份有限公司 | System theme color changing method, device and medium |
| CN110795858B (en) * | 2019-11-06 | 2023-04-07 | 广东博智林机器人有限公司 | Method and device for generating home decoration design drawing |
| CN110807561A (en) * | 2019-11-13 | 2020-02-18 | 吉林农业大学 | Bayesian network-based corn pest and disease early warning analysis method |
| CN111079001A (en) * | 2019-11-26 | 2020-04-28 | 贝壳技术有限公司 | Decoration recommendation information generation method and device, storage medium and electronic equipment |
| CN111291435A (en) * | 2020-02-27 | 2020-06-16 | 乐意节能装饰材料(常州)有限公司 | Color matching method |
| CN111429539B (en) * | 2020-03-24 | 2023-06-13 | 上海阿几网络技术有限公司 | Color matching constraint method based on color matching engine, storage medium and terminal |
| CN111400924B (en) * | 2020-03-24 | 2023-07-07 | 上海阿几网络技术有限公司 | Automatic color matching method based on color matching engine, storage medium and terminal |
| CN113298105A (en) * | 2020-07-22 | 2021-08-24 | 阿里巴巴集团控股有限公司 | Image processing, displaying, acquiring and home decoration matching processing method, device and medium |
| CN112131645B (en) * | 2020-09-23 | 2021-08-31 | 贝壳找房(北京)科技有限公司 | Indoor article placement completion method, device and equipment, and storage medium |
| CN112489230B (en) * | 2020-12-09 | 2023-03-24 | 贝壳技术有限公司 | Color matching method and device for patch model and storage medium |
| CN112652030B (en) * | 2020-12-11 | 2023-09-19 | 浙江工商大学 | Color space position layout recommendation method based on specific scene |
| KR102330823B1 (en) * | 2021-02-03 | 2021-11-24 | 구본웅 | A System Providing Furniture Recommendation Service based on Augmented Reality |
| CN113433837B (en) * | 2021-06-15 | 2023-03-21 | 浙江水利水电学院 | Indoor design method and system based on VR |
| CN113255052B (en) * | 2021-07-09 | 2021-09-24 | 佛山市陶风互联网络科技有限公司 | Home decoration scheme recommendation method and system based on virtual reality and storage medium |
| CN113449135B (en) | 2021-08-31 | 2021-11-19 | 阿里巴巴达摩院(杭州)科技有限公司 | Image generation system and method |
| CN114462207B (en) * | 2022-01-07 | 2023-03-14 | 广州极点三维信息科技有限公司 | Matching method, system, equipment and medium for home decoration template |
| CN114944101B (en) * | 2022-05-14 | 2023-04-14 | 西南医科大学 | Psychology auxiliary device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104981830A (en) * | 2012-11-12 | 2015-10-14 | 新加坡科技设计大学 | Clothing matching system and method |
| CN105389550A (en) * | 2015-10-29 | 2016-03-09 | 北京航空航天大学 | Remote sensing target detection method based on sparse guidance and significant drive |
| KR20160031873A (en) * | 2014-09-15 | 2016-03-23 | 연세대학교 산학협력단 | Gesture recognizer setting method, gesture recognition method and apparatus |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103077555B (en) * | 2013-01-18 | 2016-01-20 | 南京大学 | The automatic marking method that a kind of three-dimensional model is formed |
| CN104050722B (en) * | 2014-06-06 | 2016-12-07 | 北京航空航天大学 | The indoor three-dimensional scenic layout that a kind of picture material drives generates method with color transfer |
-
2016
- 2016-07-05 CN CN201610525746.3A patent/CN106202352B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104981830A (en) * | 2012-11-12 | 2015-10-14 | 新加坡科技设计大学 | Clothing matching system and method |
| KR20160031873A (en) * | 2014-09-15 | 2016-03-23 | 연세대학교 산학협력단 | Gesture recognizer setting method, gesture recognition method and apparatus |
| CN105389550A (en) * | 2015-10-29 | 2016-03-09 | 北京航空航天大学 | Remote sensing target detection method based on sparse guidance and significant drive |
Non-Patent Citations (1)
| Title |
|---|
| Bayesian modelling of impact of colour on web credibility;Eleftherios Papachristos et al;《17th European Conference on Artificial Intelligence》;20061231;第141卷;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106202352A (en) | 2016-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106202352B (en) | The method of indoor furniture style and colour match design based on Bayesian network | |
| US10140652B2 (en) | Computer implemented methods and systems for generating virtual body models for garment fit visualisation | |
| CN105118082B (en) | Individualized video generation method and system | |
| EP4184414A1 (en) | Product recommendation device and method based on image database analysis | |
| US20160026926A1 (en) | Clothing matching system and method | |
| CN101477560A (en) | Clothing matching system and method based on scheme database | |
| Yu et al. | DressUp!: outfit synthesis through automatic optimization. | |
| US20220366675A1 (en) | Apparatus and method for developing style analysis model based on data augmentation | |
| CN105809742B (en) | A kind of indoor scene color method of data-driven | |
| Stamps III | Simulating designed environments | |
| Podlasov et al. | Japanese street fashion for young people: A multimodal digital humanities approach for identifying sociocultural patterns and trends | |
| KR20200105507A (en) | Method and apparatus for characterizing and determining the relationship between an article and an instant | |
| Lin et al. | Example‐based image recoloring in an indoor environment | |
| KR20200122178A (en) | apparatus and method for generating movie contents components designs based on images and text | |
| KR20200122177A (en) | apparatus and method for generating game contents components designs based on images and text | |
| KR20200122179A (en) | apparatus and method for generating interior designs based on images and text | |
| CN117033688B (en) | Character image scene generation system based on AI interaction | |
| Li et al. | Controllable automatic generation of non‐player characters in 3D anime style | |
| Aziz et al. | Feminine Beauty in the Hyperreal World: A Postmodern Analysis of Esquire’s Women We Love | |
| Sugiyama et al. | Texture Synthesis Based on Aesthetic Texture Perception Using CNN Style and Content Features | |
| CN117710622A (en) | Garment design optimization color matching system based on Internet | |
| HUANG | PROBLEM SOLVING BEHAVIOR EMPLOYED IN APARTMENT INTERIOR WORKS DESIGN USING INTERACTIVE EVOLUTIONARY COMPUTATION | |
| HUANG | PROBLEM SOLVING BEHAVIOR EMPLOYED IN |
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 |