CN108537855A - A kind of ceramic marble paper method for generating pattern and device that sketch is consistent - Google Patents
A kind of ceramic marble paper method for generating pattern and device that sketch is consistent Download PDFInfo
- Publication number
- CN108537855A CN108537855A CN201810281322.6A CN201810281322A CN108537855A CN 108537855 A CN108537855 A CN 108537855A CN 201810281322 A CN201810281322 A CN 201810281322A CN 108537855 A CN108537855 A CN 108537855A
- Authority
- CN
- China
- Prior art keywords
- marble paper
- ceramic marble
- paper pattern
- ceramic
- lines
- 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.)
- Granted
Links
- 239000004579 marble Substances 0.000 title claims abstract description 293
- 239000000919 ceramic Substances 0.000 title claims abstract description 288
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 43
- 238000013473 artificial intelligence Methods 0.000 claims abstract description 19
- 238000013528 artificial neural network Methods 0.000 claims abstract description 12
- 238000012545 processing Methods 0.000 claims description 32
- 230000009466 transformation Effects 0.000 claims description 31
- 238000005070 sampling Methods 0.000 claims description 20
- 238000006243 chemical reaction Methods 0.000 claims description 15
- 238000000605 extraction Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 12
- 229910052573 porcelain Inorganic materials 0.000 claims description 11
- 230000008859 change Effects 0.000 claims description 9
- 238000003475 lamination Methods 0.000 claims description 8
- 238000000844 transformation Methods 0.000 claims description 8
- 239000000203 mixture Substances 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 4
- PXFBZOLANLWPMH-UHFFFAOYSA-N 16-Epiaffinine Natural products C1C(C2=CC=CC=C2N2)=C2C(=O)CC2C(=CC)CN(C)C1C2CO PXFBZOLANLWPMH-UHFFFAOYSA-N 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 claims description 3
- 238000011156 evaluation Methods 0.000 claims description 3
- 230000010365 information processing Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 12
- 238000010586 diagram Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 4
- 238000009825 accumulation Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 241000607479 Yersinia pestis Species 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000686 essence Substances 0.000 description 1
- 230000008921 facial expression Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/003—Reconstruction from projections, e.g. tomography
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
- G06T3/14—Transformations for image registration, e.g. adjusting or mapping for alignment of images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention belongs to technical field of information processing and artificial intelligence field more particularly to a kind of ceramic marble paper method for generating pattern and device that sketch is consistent, including：（1）Build ceramic marble paper pattern line marking model；（2）The lines for being obtained ceramic marble paper pattern using ceramic marble paper pattern line marking model according to ceramic marble paper pattern library are marked；（3）Build a kind of ceramic marble paper pattern generation model of artificial neural network approximate modeling；（4）Ceramic marble paper pattern generation model is obtained by the training method based on artificial intelligence according to ceramic marble paper pattern and its lines mark.The generation demand to ceramic marble paper pattern that a kind of consistent ceramic marble paper method for generating pattern of sketch disclosed by the invention is described by ceramic marble paper pattern generation Model Identification user's cartographical sketching, to automatically generate the ceramic marble paper pattern that meets user view consistent with user's cartographical sketching.
Description
Technical field
The invention belongs to technical field of information processing and artificial intelligence field more particularly to a kind of Ceramic flowers that sketch is consistent
Paper pattern generation method and device.
Background technology
With the development of artificial intelligence, the proposition of confrontation neural network is especially generated, computer generation can match in excellence or beauty true
The ability of picture has obtained great promotion.Such as the pest and disease damage Picture Generation Method that proposes of patent CN 107016406 A and
The human face expression Picture Generation Method that patent CN 107220600 A are proposed, the user demand that can be understood according to machine generate from
Meet the picture of user demand to a certain extent.
Cartographical sketching is mankind's one kind certainly however directly thinking alienation and exchange way, sketch can use simple shape
Shape is intended to express the thinking that people are abstracted, and in generation demand of the expression to ceramic marble paper pattern, user provides one sometimes
Cartographical sketching or the sketch of other forms may be easy intuitive way.The purpose of current sketch recognition is to utilize Freehandhand-drawing
Sketch is retrieved from image data base for the image for including sketch feature, such as patent CN 102339306 B or patent CN
Freehand Sketchy Graphics Recognition disclosed in 107633065 A of image search method and patent CN of the propositions such as 107220277 A
Deng being not present in pattern in image data base originally to be generated using user's cartographical sketching, how machine to be allowed accurately to obtain
The described demand to ceramic marble paper pattern of sketch of family offer and its corresponding parameter information are provided, automatically generated and user
It is intended to consistent ceramic marble paper pattern, is an extremely important technical problem, there is no preferable method at present.
In conclusion problem of the existing technology is：
The purpose of current sketch recognition is to retrieve for comprising sketch feature from image data base using cartographical sketching
Image, in order to allow machine to understand user's cartographical sketching and automatically generate the pattern being not present in originally in image data base, how
Machine is allowed accurately to obtain the described demand to ceramic marble paper pattern of sketch of user's offer and its corresponding parameter information,
And the ceramic marble paper pattern consistent with user view is automatically generated, it is an extremely important technical problem, there is no at present preferably
Method.
Solve the difficulty and meaning of above-mentioned technical problem：
The image of any given pixel resolution ratio is all a data point in dimensional images space, all approximate true clear
Image construction be immense dimensional images space a compact distribution it is popular（Manifold）, it is refreshing using confrontation is generated
Learning training method through network can make the data that computer samples in the image space of higher-dimension be substantially distributed in the stream
On row, present approach provides the ceramic marble paper pattern generation models that a kind of cartographical sketching inspires so that computer can be
It is sampled in specific distribution space in the prevalence, to generate the approximate true ceramic marble paper pattern consistent with cartographical sketching.
Invention content
In view of the problems of the existing technology, the present invention provides a kind of ceramic marble paper method for generating pattern that sketch is consistent
And device.
The invention is realized in this way the consistent ceramic marble paper method for generating pattern of the sketch includes the following steps：
Step 1: structure ceramic marble paper pattern line marking model；
Step 2: according to ceramic marble paper pattern library, ceramic marble paper pattern is obtained using ceramic marble paper pattern line marking model
Lines mark；
Step 3: building a kind of ceramic marble paper pattern generation model of artificial neural network approximate modeling；
Step 4: obtaining ceramic marble paper by the training method based on artificial intelligence according to ceramic marble paper pattern and its lines mark
Pattern generation model.
In step 1, the ceramic marble paper pattern line marking model includes two processing modules：Pattern edge carries
Modulus block and lines figure conversion module；
The pattern edge extraction module extracts image border etc. by the Boundary extracting algorithm in Digital Image Processing and processes
Cheng Shixian；
The lines figure conversion module handled by digital image in the graph transformations such as deformation, distortion and simplify transformation；Institute
Stating the transformation such as deformation, distortion can be realized by similarity transformation, affine transformation or the projective transformation in Digital Image Processing；It is described
Simplifying transformation can be realized by Douglas-Pu Ke evacuating algorithms.
In step 2, ceramic marble paper pattern is generated and ceramic marble paper pattern through ceramic marble paper pattern line marking model
Corresponding lines annotation process is：
Ceramic marble paper pattern is inputted to the pattern edge extraction module of porcelain marble paper pattern line marking model, obtains the pottery with input
The corresponding initial lines figure of porcelain marble paper pattern；
The lines figure conversion module that the initial lines figure of ceramic marble paper pattern is inputted to porcelain marble paper pattern line marking model, obtains line
Lines after the ceramic marble paper pattern change of lines figure composition of the item figure after the graph transformations such as deforming, distorting and simplify transformation
Figure；
By lines figure after the initial lines figure of ceramic marble paper pattern and ceramic marble paper pattern change collectively as ceramic marble paper pattern
Lines mark collection.
In the corresponding lines mark collection of structure ceramic marble paper pattern, by mark user's Freehandhand-drawing and ceramic marble paper is also collected
The corresponding lines figure of pattern, and ceramic marble paper figure is added after the processing of lines figure conversion module in the lines figure for marking user's Freehandhand-drawing
The lines mark collection of case.
In step 3, carry out approximate modeling ceramic marble paper using the artificial neural network of the combo architectures comprising a variety of hidden layers
Pattern generation model step；
The ceramic marble paper pattern generation model includes input layer, the first convolutional layer, the second convolutional layer, third convolutional layer,
One residual error network block layer, the first up-sampling layer, the second up-sampling layer, output layer, using the cartographical sketching that user provides as ceramics
The input layer of marble paper pattern generation model, the input is by first convolutional layer, second convolutional layer, third volume
Lamination, the first residual error network block layer, it is described first up-sampling layer processing after, it is described first up-sampling layer output data and
The output data of second convolutional layer up-samples layer processing, the output data and the of the second up-sampling layer through described second jointly
The output data of one convolutional layer jointly through the output layer processing after output data；
In step 3, the training process that ceramic marble paper pattern generation model is obtained by training is by artificial intelligence
Practise the output data energy minimization first-loss function that training makes ceramic marble paper pattern generation model；
The first-loss function is defined as follows：
Wherein,Indicate the output data of the ceramic marble paper pattern generation model,Sentenced by ceramic marble paper pattern
Other model definition,It is ceramic marble paper pattern to enable ceramic marble paper pattern discrimination model to distinguish input
Training sample,To enable the output data of ceramic marble paper pattern generation model to cheat pottery
Porcelain marble paper pattern discrimination model,To make the defeated of ceramic marble paper pattern generation model
The characteristics of image for going out data is consistent with the characteristics of image of ceramic marble paper pattern training sample,It indicates in ceramic marble paper figure
The of case discrimination modelLayer extraction feature.
The ceramic marble paper pattern discrimination model is by convolutional neural networks approximate modeling, ceramic marble paper pattern discrimination model packet
Input layer, the first convolutional layer, the second convolutional layer, third convolutional layer, Volume Four lamination, full articulamentum are included, institute is passed through in the input
It states input layer, first convolutional layer, second convolutional layer, the third convolutional layer, the Volume Four lamination, described connect entirely
Connect output data after layer；
Another object of the present invention is to provide a kind of ceramic marble paper pattern generating apparatus that sketch is consistent, the consistent potteries of the sketch
Porcelain marble paper pattern generating apparatus includes：
First generation module：Lines for obtaining ceramic marble paper pattern according to ceramic marble paper pattern library mark；
Second generation module：For obtaining pottery by the training method of artificial intelligence according to ceramic marble paper pattern and its lines mark
Porcelain marble paper pattern generation model；
First processing module：For receiving the generation demand input by user expressed with lines sketch form；
First acquisition module：For by ceramic marble paper pattern generation model by the lines sketch and random perturbation that currently input with
The lines mark fusion of history input, and obtain the ceramic marble paper pattern consistent with user demand；
Second processing module：For showing the ceramic marble paper pattern currently generated by ceramic marble paper pattern generation model, and receive
Selection and evaluation of the user to ceramic marble paper pattern；
Third processing module：Ceramic marble paper pattern for selecting user and input by user with the expression of lines sketch form
Generation demand is combined marks sample to obtain new ceramic marble paper pattern line.
In conclusion advantages of the present invention and good effect are：
1. the method for the present invention is marked using pattern edge extraction module and lines figure conversion module composition ceramic marble paper pattern line
Model, machine can by the model obtain ceramic marble paper pattern lines mark, efficiently solve training sample it is few and obtain at
This high problem；
2. the method for the present invention for user's cartographical sketching to ceramic marble paper pattern mapping feature devise it is a kind of completely newly include
The artificial neural network of a variety of hidden layer frameworks, and ceramic marble paper pattern generation mould is obtained by the learning training method of artificial intelligence
Type；The generation demand to ceramic marble paper pattern described by ceramic marble paper pattern generation Model Identification user's cartographical sketching, and
The ceramic marble paper pattern consistent with user's cartographical sketching is generated, the ceramic marble paper pattern consistent with cartographical sketching is efficiently solved
Machine automatically generates problem.
Description of the drawings
Fig. 1 is a kind of flow chart of the consistent ceramic marble paper method for generating pattern of sketch according to an embodiment of the invention；
Fig. 2 is the lines marking structure schematic diagram according to an embodiment of the invention for obtaining ceramic marble paper pattern；
Fig. 3 is the structural schematic diagram that ceramic marble paper according to an embodiment of the invention generates model；
Fig. 4 is the structural schematic diagram of ceramic marble paper pattern discrimination model according to an embodiment of the invention；
Fig. 5 is the ceramic marble paper method for generating pattern of the sketch recognition according to an embodiment of the invention based on artificial intelligence
Flow chart；
Fig. 6 is the ceramic marble paper pattern generation side of the sketch recognition based on artificial intelligence according to another embodiment of the invention
The flow chart of method；
Fig. 7 is the ceramic marble paper pattern generating apparatus of the sketch recognition according to an embodiment of the invention based on artificial intelligence
Structural schematic diagram；
Fig. 8 is the ceramic marble paper pattern generation dress of the sketch recognition based on artificial intelligence according to another embodiment of the invention
The structural schematic diagram set.
Specific implementation mode
In order to make the purpose , technical scheme and advantage of the present invention be clearer, with reference to embodiments, the present invention is carried out
It is further described.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to limit
The present invention.
Below with reference to the accompanying drawings the ceramic marble paper pattern life of the sketch recognition based on artificial intelligence of the embodiment of the present invention is described
At method and apparatus.
As shown in Figure 1, the ceramic marble paper pattern generation demand recognition methods packet based on sketch recognition of the embodiment of the present invention
Include following steps：
Step S101 builds ceramic marble paper pattern line marking model；
In this step, structure ceramic marble paper pattern line marking model is in order to which machine is automatically to ceramic marble paper pattern progress lines
Mark, is described in detail as follows in conjunction with Fig. 2：
In this step, the ceramic marble paper pattern line marking model 200 includes two processing modules：Pattern edge extracts
Module 202 and lines figure conversion module 203；
The pattern edge extraction module 202 extracts the processing such as image border by the Boundary extracting algorithm in Digital Image Processing
Process is realized；
The lines figure conversion module 203 handled by digital image in the graph transformations such as deformation, distortion and simplify transformation；
The transformation such as the deformation, distortion can be realized by similarity transformation, affine transformation or the projective transformation in Digital Image Processing；Institute
It states to simplify and converts and can be realized by Douglas-Pu Ke evacuating algorithms.
Step S102 obtains ceramic marble paper according to ceramic marble paper pattern library using ceramic marble paper pattern line marking model
The lines of pattern mark；
In this step, when giving ceramic marble paper pattern, approach there are two types of the lines marks of the ceramic marble paper pattern is obtained：When
By lines corresponding with the ceramic marble paper pattern mark for marking user's Freehandhand-drawing；Second is that marking mould by ceramic marble paper pattern line
Type inputs ceramic marble paper pattern, and automatically obtaining lines corresponding with ceramic marble paper pattern by machine marks.It is detailed in conjunction with Fig. 2
It is described as follows：
As shown in Fig. 2, ceramic marble paper pattern 201 is inputted pattern edge extraction module 202, the ceramic marble paper figure with input is obtained
The corresponding initial lines figure of case；
As shown in Fig. 2, by the initial lines figure input lines figure conversion module 203 of ceramic marble paper pattern, lines figure is obtained by becoming
Lines figure after the ceramic marble paper pattern change of lines figure composition after the graph transformations such as shape, distortion and simplified transformation；
As shown in Fig. 2, by lines figure after the initial lines figure of ceramic marble paper pattern and ceramic marble paper pattern change collectively as ceramics
The lines mark collection 204 of marble paper pattern.
Preferably, it at corresponding lines mark collection 204 of structure ceramic marble paper pattern, also collects by marking user's Freehandhand-drawing
Lines figure corresponding with ceramic marble paper pattern, and the lines figure of user's Freehandhand-drawing will be marked after the processing of lines figure conversion module 203
The lines mark collection 304 of ceramic marble paper pattern is added.
Further, the ceramic marble paper pattern and its lines annotation definition that this step is obtained are as follows：
Wherein,Indicate theA ceramic marble paper pattern,Indicate theA ceramic marble paper pattern
IncludeItem mark（Attribute or sentence）Mark collection,Indicate the of the mark collectionItem mark.
Step S103 builds a kind of ceramic marble paper pattern generation model of artificial neural network approximate modeling；
In this step, carrys out approximate modeling ceramic marble paper pattern using the artificial neural network of the combo architectures comprising a variety of hidden layers and give birth to
At model, in conjunction with Fig. 3, detailed description are as follows：
As shown in figure 3, the input data of ceramic marble paper pattern generation model is the lines mark of ceramic marble paper pattern, by Ceramic flower
The lines of paper pattern mark the input layer 301 as ceramic marble paper pattern generation model.
It is handled as shown in figure 3, input data inputs the first convolutional layer 302 via the input layer 301.
Specifically, first convolutional layer 302 includes the convolution algorithm that convolution step-length is 1, and the quantity of convolution kernel is 64, volume
The size of product core is 7*7.
As shown in figure 3, after input data is handled via first convolutional layer 302, input the second convolutional layer 303 processing.
Specifically, second convolutional layer 303 includes the convolution algorithm that convolution step-length is 2, and the quantity of convolution kernel is 128,
The size of convolution kernel is 5*5.
As shown in figure 3, after input data is handled via second convolutional layer 303, input third convolutional layer 304 is handled.
Specifically, the third convolutional layer 304 includes the convolution algorithm that convolution step-length is 2, and the quantity of convolution kernel is 256,
The size of convolution kernel is 5*5.
As shown in figure 3, after data are handled via third convolutional layer 304, input the first residual error network block layer 305 processing.
Specifically, the first residual error network block layer 305 includes A residual error network block（ResNet Block）306；It is excellent
Choosing, between the value of A is 5~9.
As shown in figure 3, after data are handled via the first residual error network block layer 305, input the first up-sampling layer 307 is handled.
Preferably, the first up-sampling layer 307 can be by sub-pix convolution（Sub-pixel Convolution）Layer is real
It is existing；Optionally, the first up-sampling layer 307 can be by deconvolution（Deconvolution）Layer is realized.
As shown in figure 3, after data are handled via the first up-sampling layer 307, input the second up-sampling layer 308 is handled.
Specifically, there are two sources for the input data of the second up-sampling layer 308, first, the first up-sampling layer
307 output data, second is that the output data of the second convolutional layer 303 by fast connecting 309 transmission.
As shown in figure 3, the second up-sampling layer 308 can be by sub-pix convolution（Sub-pixel Convolution）
Layer is realized；Optionally, the second up-sampling layer 308 can be by deconvolution（Deconvolution）Layer is realized
As shown in figure 3, after data are handled via the second up-sampling layer 308, by the output layer 311 of ceramic marble paper pattern generation model
Output.
Specifically, the output layer 311 includes the convolution algorithm that convolution step-length is 1, and the quantity of convolution kernel is 3, convolution kernel
Size be 3*3, output layer 311 the result is that triple channel real number matrix, i.e. chromatic ceramics marble paper pattern.It should be noted that
The convolutional layer of ceramic marble paper pattern generation model does not include pond（Pooling）Operation.
Step S104 obtains pottery according to ceramic marble paper pattern and its lines mark by the training method based on artificial intelligence
Porcelain marble paper pattern generation model.
In this step, the learning training process that ceramic marble paper pattern generation model is obtained by training is to introduce ceramics
Marble paper pattern discrimination model, ceramic marble paper pattern discrimination model carrys out approximate modeling using convolutional neural networks, detailed in conjunction with Fig. 4
It is described as follows：
As shown in figure 4, the input of input layer 401 of ceramic marble paper pattern discrimination model is a real number matrix, the first convolutional layer
402 make the convolution kernel of 64 5*5, convolution step-length be 2, and the second convolutional layer 403 makes the convolution kernel of 128 5*5, convolution step-length be 2,
Third convolutional layer 404 makes the convolution kernel of 256 5*5, convolution step-length be 2, and Volume Four lamination 405 makes the convolution kernel of 512 5*5,
Convolution step-length is 2, and the first full articulamentum 406 includes a neuron, is connect entirely with Volume Four lamination.
It should be noted that each convolutional layer as shown in Figure 4 includes normalization operation and line rectification operation, but not
Including pond operation.
Further, in order to make those skilled in the art be more clear how by training obtain ceramic marble paper pattern semanteme sketch
Model is generated, specific training process is as follows：
The first step, initial phase, the parameter matrix of each hidden layer of random initializtion ceramic marble paper pattern generation model are random first
The parameter matrix of each hidden layer of beginningization ceramic marble paper pattern discrimination model；
Ceramic marble paper pattern line is marked the input as ceramic marble paper pattern generation model by second step, propagated forward stage
Layer 301, data obtain output data after the transformation of each hidden layer, in the output layer 311 of ceramic marble paper pattern generation model；
Third walks, training ceramic marble paper pattern discrimination model so that the input layer 401 of the ceramic marble paper pattern discrimination model is defeated
When entering the data that the output layer 311 that data are ceramic marble paper pattern generation models exports, the ceramic marble paper pattern discrimination model
The first full articulamentum 406 output be 0；When 401 input data of ceramic marble paper pattern discrimination model input layer is given
When ceramic marble paper pattern, the first full articulamentum 406 output of the ceramic marble paper pattern discrimination model is 1, i.e., so that Ceramic flower
The second loss function of output energy minimization of paper pattern discrimination model；
Specifically, second loss function is analytic function of the first-loss function in training image discrimination model, tool
Body is defined as：
Wherein,Indicate the ceramic marble paper pattern of training sample,Indicate the input layer 301 of ceramic marble paper pattern generation model
Input data,Indicate the output data of the output layer 311 of ceramic marble paper pattern generation model.
4th step, back-propagation phase, calculate ceramic marble paper pattern generation model output layer 311 obtain output data with
The error of ideal output, it is reversed to adjust each hidden of ceramic marble paper pattern generation model by the method for minimizing third loss function
The parameter matrix of layer.
Specifically, the third loss function is first-loss function in training ceramic marble paper pattern generation model
When analytic function, be specifically defined as：
Wherein,Indicate the input data of the input layer 301 of ceramic marble paper pattern generation model,Indicate output layer 311
Output data,Indicate the given ceramic marble paper pattern of training sample,It is normalization factor,By having instructed
The image discriminating model perfected is defined as, willAs the input data of image discriminating model, the of image discriminating model
The characteristics of image of convolutional layer extraction, then cumulative itemIt can be described as：Calculate ceramic marble paper figure
Case generates the output data of the output layer 311 of model, the of image discriminating modelThe image of convolutional layer extraction is special
The given ceramic marble paper pattern of training sample of seeking peace, the of image discriminating modelThe characteristics of image of convolutional layer extraction
Euclidean distance.
Step 5 repeats second step to the 4th step, until the value of error function is less than a given threshold value or changes
Generation number reaches a given threshold value.
It should be noted that those skilled in the art will appreciate that the training process of the ceramic marble paper pattern generation model is one
The training process for kind generating confrontation, can also implement above-mentioned to train with the training method and technology for generating confrontation neural network
Journey.
In conclusion the ceramic marble paper method for generating pattern that a kind of sketch of the embodiment of the present invention is consistent, first according to pottery
Porcelain marble paper pattern library obtains ceramic marble paper pattern line mark, then according to ceramic marble paper pattern and corresponding ceramic marble paper pattern
Lines mark obtains ceramic marble paper pattern generation model according to the training method of artificial intelligence.Gao Shi is generated by training as a result,
The ceramic marble paper pattern generation model of answering property and high automation, the Ceramic flower expressed with sketch form that can be provided according to user
Paper pattern generates demand and generates ceramic marble paper pattern consistent therewith via ceramic marble paper pattern generation model, improves Ceramic flower
Paper pattern generates the goodness of fit with user view.
As shown in figure 5, the ceramic marble paper pattern generation side that a kind of sketch according to another embodiment of the invention is consistent
The flow chart of method.
Embodiment based on Fig. 1 further includes after step s 104：
Step S501 obtains the ceramic marble paper pattern line sketch for the Freehandhand-drawing that user provides；
Step S502 by the lines sketch provided corresponding to user and samples the random noise from normal distribution together as defeated
Enter, inputs the input layer of ceramic marble paper pattern generation model；
Step S503, by the hidden layer of ceramic marble paper pattern generation model by the lines sketch currently inputted and random noise with go through
The lines mark of history input carries out fusion calculation；
Step S504 is obtained corresponding to user's offer in the output layer of ceramic marble paper pattern generation model with lines sketch form
The ceramic marble paper pattern of the generation demand of expression.
In conclusion the ceramic marble paper method for generating pattern of the sketch recognition based on artificial intelligence of the embodiment of the present invention,
The generation demand expressed with lines sketch form and random perturbation noise the input ceramic marble paper pattern life that user is provided first
At model, then the hidden layer of ceramic marble paper pattern generation model inputs the lines sketch and random noise that currently input and history
Lines mark merged, then the output layer of ceramic marble paper pattern generation model obtain correspond to user provide with line
The ceramic marble paper pattern of the generation demand of sketch form expression.
As shown in fig. 6, the ceramic marble paper pattern generation side that a kind of sketch according to still a further embodiment is consistent
The flow chart of method.
Based on above-described embodiment, in order to further increase the generation of ceramic marble paper pattern generation model ceramic marble paper pattern and
The goodness of fit of user demand, ceramic marble paper pattern generation model generate several ceramics according to the lines sketch and random perturbation of input
Marble paper pattern, user can select the ceramic marble paper pattern most identical with its demand, user provide with lines sketch form table
The generation demand reached and the ceramic marble paper pattern of its selection constitute the lines mark sample of new ceramic marble paper pattern, and in sample
Quantity when running up to certain threshold value, re -training ceramic marble paper pattern generation model, detailed process is described as follows：
Step S601 obtains the ceramic marble paper pattern of user's selection, and the ceramic marble paper pattern of user's selection and expression user are needed
The lines sketch combination asked constitutes new ceramic marble paper pattern line mark sample；
When the quantity of step S602, the new ceramic marble paper pattern line mark sample of accumulation reach predetermined threshold value, re -training
Ceramic marble paper pattern generation model.
It should be noted that the new ceramic marble paper pattern line mark sample size in accumulation is not up to predetermined threshold value
When, not re -training ceramic marble paper pattern generation model.
As shown in fig. 7, a kind of consistent ceramic marble paper pattern generating apparatus of sketch according to an embodiment of the invention
Structure, including：
First generation module 710：Lines for obtaining ceramic marble paper pattern according to ceramic marble paper pattern library mark.
Specifically, ceramic marble paper pattern is inputted into pattern edge extraction module, obtains the ceramic marble paper pattern pair with input
The initial lines figure answered；Then the initial lines figure of ceramic marble paper pattern is inputted into lines figure conversion module, obtains lines figure and passes through
Lines figure after the ceramic marble paper pattern change of lines figure composition after the graph transformations such as deformation, distortion and simplified transformation；Finally
Lines mark of the lines figure collectively as ceramic marble paper pattern after the initial lines figure of ceramic marble paper pattern and ceramic marble paper pattern change
Note.Meanwhile the lines figure for marking user's Freehandhand-drawing is also inputted into lines figure conversion module, it obtains mark user's free hand line figure and passes through
The graph transformations such as deformation, distortion and simplify the lines figure after transformation, these lines figures also lines as ceramic marble paper pattern
Mark.
Second generation module 720：Training side for passing through artificial intelligence according to ceramic marble paper pattern and its lines mark
Method obtains ceramic marble paper pattern generation model.
Specifically, image discriminating model, the ceramic marble paper pattern are introduced when training ceramic marble paper pattern generation model
It generates model and image discriminating model carrys out approximate modeling using artificial neural network, in the error function described according to the present invention
Under conditions of, the methods and techniques of training ceramic marble paper pattern generation model generate the training one of confrontation neural network with condition
It causes.
It should be noted that a kind of explanation for the ceramic marble paper method for generating pattern embodiment that sketch is consistent of aforementioned base
It is also applied for a kind of consistent ceramic marble paper pattern generating apparatus of sketch of the embodiment, realization principle is similar, herein no longer
It repeats.
In conclusion the ceramic marble paper pattern generating apparatus that a kind of sketch of the embodiment of the present invention is consistent, first according to pottery
Porcelain marble paper pattern library obtains ceramic marble paper pattern line mark, passes through people then according to ceramic marble paper pattern line labeled data collection
The training method of work intelligence obtains ceramic marble paper pattern generation model.It is automatic by the high-adaptability and height of training generation as a result,
The ceramic marble paper pattern generation model of change, the generation demand with the expression of lines sketch form that can be provided according to user are accurately compiled
More attribute requirements of code user, improve the goodness of fit of ceramic marble paper pattern generation and user view.
As shown in Figure 8, according to another embodiment of the invention a kind of consistent ceramic marble paper pattern generation dress of sketch
The structure set, including：
First processing module 830：For receiving the generation demand input by user expressed with lines sketch form.
First acquisition module 840：For by ceramic marble paper pattern generation model by the lines sketch currently inputted and with
Machine disturbs the lines mark inputted with history and merges, and obtains the ceramic marble paper pattern consistent with user demand.
Second processing module 850：For showing the ceramic marble paper pattern currently generated by ceramic marble paper pattern generation model,
And receive selection and evaluation of the user to ceramic marble paper pattern.
Third processing module 860：Ceramic marble paper pattern for selecting user and input by user with lines sketch shape
The generation demand of formula expression is combined marks sample to obtain new ceramic marble paper pattern line.
It should be noted that a kind of explanation of the consistent ceramic marble paper method for generating pattern embodiment of aforementioned sketch
A kind of consistent ceramic marble paper pattern generating apparatus of sketch suitable for the embodiment, realization principle are similar.
In conclusion the ceramic marble paper pattern generating apparatus that the sketch of the embodiment of the present invention is consistent, it is defeated to receive user first
The ceramic marble paper pattern generation demand expressed with lines sketch form entered then needs the generation expressed with lines sketch form
Summation random perturbation inputs ceramic marble paper pattern generation model together, and will be worked as by the hidden layer of ceramic marble paper pattern generation model
The lines sketch and random perturbation of preceding input are merged with the lines mark that history inputs, finally in ceramic marble paper pattern generation model
Output layer obtain ceramic marble paper pattern.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all essences in the present invention
All any modification, equivalent and improvement etc., should all be included in the protection scope of the present invention made by within refreshing and principle.
Claims (8)
1. a kind of ceramic marble paper method for generating pattern that sketch is consistent, which is characterized in that the consistent ceramic marble paper figure of the sketch
Case generation method includes the following steps：
Step 1: building a kind of ceramic marble paper pattern line marking model；
Step 2: according to ceramic marble paper pattern library, ceramic marble paper pattern is obtained using ceramic marble paper pattern line marking model
Lines mark；
Step 3: building a kind of ceramic marble paper pattern generation model of artificial neural network approximate modeling；
Step 4: obtaining ceramic marble paper by the training method based on artificial intelligence according to ceramic marble paper pattern and its lines mark
Pattern generation model.
2. the consistent ceramic marble paper method for generating pattern of sketch as described in claim 1, which is characterized in that in step 1,
The ceramic marble paper pattern line marking model includes pattern edge extraction module and lines figure conversion module；
The pattern edge extraction module extracts image border etc. by the Boundary extracting algorithm in Digital Image Processing and processes
Cheng Shixian；
The lines figure conversion module handled by digital image in the graph transformations such as deformation, distortion and simplify transformation；Institute
Stating the transformation such as deformation, distortion can be realized by similarity transformation, affine transformation or the projective transformation in Digital Image Processing；It is described
Simplifying transformation can be realized by Douglas-Pu Ke evacuating algorithms.
3. the consistent ceramic marble paper method for generating pattern of sketch as described in claim 1, it is characterised in that, will in step 2
Ceramic marble paper pattern generates lines annotation process corresponding with ceramic marble paper pattern through ceramic marble paper pattern line marking model：
Ceramic marble paper pattern is inputted into pattern edge extraction module first, is obtained corresponding with the ceramic marble paper pattern of input initial
Lines figure；Then the initial lines figure of ceramic marble paper pattern is inputted into lines figure conversion module, obtains lines figure by deformation, distortion
Lines figure after the ceramic marble paper pattern change of lines figure composition after equal graph transformations and simplified transformation；Finally by ceramic marble paper
Lines figure marks collection collectively as the lines of ceramic marble paper pattern after the initial lines figure of pattern and ceramic marble paper pattern change.
4. the consistent ceramic marble paper method for generating pattern of sketch as claimed in claim 3, which is characterized in that in structure Ceramic flower
When the corresponding lines mark of paper pattern collects, the lines figure corresponding with ceramic marble paper pattern by marking user's Freehandhand-drawing is also collected, and
The lines that the lines figure for marking user's Freehandhand-drawing is added to ceramic marble paper pattern after the processing of lines figure conversion module mark collection.
5. the consistent ceramic marble paper method for generating pattern of sketch as described in claim 1, which is characterized in that in step 3,
Carry out approximate modeling ceramic marble paper pattern generation model using the artificial neural network of the combo architectures comprising a variety of hidden layers；
The ceramic marble paper pattern generation model includes input layer, the first convolutional layer, the second convolutional layer, third convolutional layer,
One residual error network block layer, the first up-sampling layer, the second up-sampling layer, output layer, using the cartographical sketching that user provides as ceramics
The input layer of marble paper pattern generation model, the input is by first convolutional layer, second convolutional layer, third volume
Lamination, the first residual error network block layer, it is described first up-sampling layer processing after, it is described first up-sampling layer output data and
The output data of second convolutional layer up-samples layer processing, the output data and the of the second up-sampling layer through described second jointly
The output data of one convolutional layer jointly through the output layer processing after output data.
6. the consistent ceramic marble paper method for generating pattern of sketch as described in claim 1, which is characterized in that in step 4,
The training process that ceramic marble paper pattern generation model is obtained by training is by network training ceramic marble paper pattern to be given birth to
At the output data energy minimization loss function of model；
The loss function is as follows：
Wherein,Indicate the output data of the ceramic marble paper pattern generation model,Sentenced by ceramic marble paper pattern
Other model definition,It is ceramic marble paper pattern to enable ceramic marble paper pattern discrimination model to distinguish input
Training sample,To enable the output data of ceramic marble paper pattern generation model to cheat pottery
Porcelain marble paper pattern discrimination model,To make the defeated of ceramic marble paper pattern generation model
The characteristics of image for going out data is consistent with the characteristics of image of ceramic marble paper pattern training sample,It indicates in ceramic marble paper figure
The of case discrimination modelLayer extraction feature；
The ceramic marble paper pattern discrimination model is by convolutional neural networks approximate modeling, the ceramic marble paper pattern discrimination model packet
Input layer, the first convolutional layer, the second convolutional layer, third convolutional layer, Volume Four lamination, full articulamentum are included, institute is passed through in the input
It states input layer, first convolutional layer, second convolutional layer, the third convolutional layer, the Volume Four lamination, described connect entirely
Connect output data after layer.
7. a kind of ceramic marble paper figure that the sketch for the ceramic marble paper method for generating pattern that sketch as described in claim 1 is consistent is consistent
Case generating means, which is characterized in that the consistent ceramic marble paper pattern generating apparatus of the sketch includes the first generation module and the
Two generation modules；
The lines that first generation module is used to obtain ceramic marble paper pattern according to ceramic marble paper pattern library mark；
Second generation module is used to be obtained by the training method of artificial intelligence according to ceramic marble paper pattern and its lines mark
Take ceramic marble paper pattern generation model.
8. the consistent ceramic marble paper pattern generating apparatus of sketch as claimed in claim 7, which is characterized in that the consistent pottery of sketch
Porcelain marble paper pattern generating apparatus includes first processing module, the first acquisition module, Second processing module, third processing module；
The first processing module is for receiving the generation demand input by user expressed with lines sketch form；
First acquisition module is used to disturb by the lines sketch currently inputted and at random by ceramic marble paper pattern generation model
The dynamic lines mark inputted with history merges, and obtains the ceramic marble paper pattern consistent with user demand；
The Second processing module connects for showing the ceramic marble paper pattern currently generated by ceramic marble paper pattern generation model
Receive selection and evaluation of the user to ceramic marble paper pattern；
The third processing module is used for the ceramic marble paper pattern that selects user and input by user with lines sketch form table
The generation demand reached is combined marks sample to obtain new ceramic marble paper pattern line.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810281322.6A CN108537855B (en) | 2018-04-02 | 2018-04-02 | Ceramic stained paper pattern generation method and device with consistent sketch |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810281322.6A CN108537855B (en) | 2018-04-02 | 2018-04-02 | Ceramic stained paper pattern generation method and device with consistent sketch |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108537855A true CN108537855A (en) | 2018-09-14 |
| CN108537855B CN108537855B (en) | 2022-04-26 |
Family
ID=63482235
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810281322.6A Active CN108537855B (en) | 2018-04-02 | 2018-04-02 | Ceramic stained paper pattern generation method and device with consistent sketch |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108537855B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109285182A (en) * | 2018-09-29 | 2019-01-29 | 北京三快在线科技有限公司 | Model generating method, device, electronic equipment and computer readable storage medium |
| CN114861247A (en) * | 2022-07-06 | 2022-08-05 | 广东时谛智能科技有限公司 | Method, device, equipment and storage medium for generating shoe body model based on simple design |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050100243A1 (en) * | 2001-11-19 | 2005-05-12 | Microsoft Corporation | Automatic sketch generation |
| US20100254594A1 (en) * | 2007-11-29 | 2010-10-07 | Tencent Technology (Shenzhen) Company Ltd. | Sketch generating system and method for generating sketch based on image |
| CN106482700A (en) * | 2016-10-12 | 2017-03-08 | 东南大学 | A kind of digitized House Property Area Surveying measuring method of sketch direct mapping |
| CN107220600A (en) * | 2017-05-17 | 2017-09-29 | 清华大学深圳研究生院 | A kind of Picture Generation Method and generation confrontation network based on deep learning |
| CN107330956A (en) * | 2017-07-03 | 2017-11-07 | 广东工业大学 | A kind of unsupervised painting methods of caricature manual draw and device |
-
2018
- 2018-04-02 CN CN201810281322.6A patent/CN108537855B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050100243A1 (en) * | 2001-11-19 | 2005-05-12 | Microsoft Corporation | Automatic sketch generation |
| US20100254594A1 (en) * | 2007-11-29 | 2010-10-07 | Tencent Technology (Shenzhen) Company Ltd. | Sketch generating system and method for generating sketch based on image |
| CN106482700A (en) * | 2016-10-12 | 2017-03-08 | 东南大学 | A kind of digitized House Property Area Surveying measuring method of sketch direct mapping |
| CN107220600A (en) * | 2017-05-17 | 2017-09-29 | 清华大学深圳研究生院 | A kind of Picture Generation Method and generation confrontation network based on deep learning |
| CN107330956A (en) * | 2017-07-03 | 2017-11-07 | 广东工业大学 | A kind of unsupervised painting methods of caricature manual draw and device |
Non-Patent Citations (1)
| Title |
|---|
| 朱穆朗玛 等: "基于道格拉斯-普克法提取55例齿痕舌图像特征研究", 《中华中医药学刊》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109285182A (en) * | 2018-09-29 | 2019-01-29 | 北京三快在线科技有限公司 | Model generating method, device, electronic equipment and computer readable storage medium |
| CN114861247A (en) * | 2022-07-06 | 2022-08-05 | 广东时谛智能科技有限公司 | Method, device, equipment and storage medium for generating shoe body model based on simple design |
| CN114861247B (en) * | 2022-07-06 | 2022-12-30 | 广东时谛智能科技有限公司 | Method, device, equipment and storage medium for generating shoe body model based on simple design |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108537855B (en) | 2022-04-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113239954B (en) | Attention mechanism-based image semantic segmentation feature fusion method | |
| CN105426919B (en) | The image classification method of non-supervisory feature learning is instructed based on conspicuousness | |
| CN108280397B (en) | Human body image hair detection method based on deep convolutional neural network | |
| CN110674741B (en) | Gesture recognition method in machine vision based on double-channel feature fusion | |
| Kadam et al. | Detection and localization of multiple image splicing using MobileNet V1 | |
| CN112801015B (en) | Multi-mode face recognition method based on attention mechanism | |
| CN107844795A (en) | Convolutional neural networks feature extracting method based on principal component analysis | |
| CN107256246A (en) | PRINTED FABRIC image search method based on convolutional neural networks | |
| CN107229757A (en) | The video retrieval method encoded based on deep learning and Hash | |
| CN107316015A (en) | A kind of facial expression recognition method of high accuracy based on depth space-time characteristic | |
| CN106778506A (en) | A kind of expression recognition method for merging depth image and multi-channel feature | |
| CN110188708A (en) | A kind of facial expression recognizing method based on convolutional neural networks | |
| CN106203356B (en) | A kind of face identification method based on convolutional network feature extraction | |
| Aghamaleki et al. | Multi-stream CNN for facial expression recognition in limited training data | |
| CN109829924B (en) | Image quality evaluation method based on principal feature analysis | |
| CN112580515B (en) | Lightweight face key point detection method based on Gaussian heat map regression | |
| CN110135277B (en) | Human behavior recognition method based on convolutional neural network | |
| CN108615228A (en) | Facial image complementing method based on hybrid neural networks | |
| CN111652273A (en) | Deep learning-based RGB-D image classification method | |
| CN108257194B (en) | Face simple stroke generation method based on convolutional neural network | |
| CN100487732C (en) | Method for generating cartoon portrait based on photo of human face | |
| CN111368637B (en) | Transfer robot target identification method based on multi-mask convolutional neural network | |
| Sui et al. | FFNet-M: Feature fusion network with masks for multimodal facial expression recognition | |
| CN106650798A (en) | Indoor scene recognition method combining deep learning and sparse representation | |
| CN108537855A (en) | A kind of ceramic marble paper method for generating pattern and device that sketch is consistent |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |