CN101604451A - A kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar - Google Patents
A kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar Download PDFInfo
- Publication number
- CN101604451A CN101604451A CNA2009101006502A CN200910100650A CN101604451A CN 101604451 A CN101604451 A CN 101604451A CN A2009101006502 A CNA2009101006502 A CN A2009101006502A CN 200910100650 A CN200910100650 A CN 200910100650A CN 101604451 A CN101604451 A CN 101604451A
- Authority
- CN
- China
- Prior art keywords
- chinese calligraphy
- stroke
- word
- certain
- confidence level
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 113
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 47
- 230000002708 enhancing effect Effects 0.000 claims abstract description 18
- 238000013528 artificial neural network Methods 0.000 claims description 31
- 238000012549 training Methods 0.000 claims description 26
- 238000005728 strengthening Methods 0.000 claims description 13
- 238000011156 evaluation Methods 0.000 claims description 9
- 238000013459 approach Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000002787 reinforcement Effects 0.000 claims description 7
- YTAHJIFKAKIKAV-XNMGPUDCSA-N [(1R)-3-morpholin-4-yl-1-phenylpropyl] N-[(3S)-2-oxo-5-phenyl-1,3-dihydro-1,4-benzodiazepin-3-yl]carbamate Chemical compound O=C1[C@H](N=C(C2=C(N1)C=CC=C2)C1=CC=CC=C1)NC(O[C@H](CCN1CCOCC1)C1=CC=CC=C1)=O YTAHJIFKAKIKAV-XNMGPUDCSA-N 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 5
- 230000006978 adaptation Effects 0.000 claims description 4
- 239000000470 constituent Substances 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 4
- 238000007418 data mining Methods 0.000 claims description 4
- 238000012938 design process Methods 0.000 claims description 4
- 230000001537 neural effect Effects 0.000 claims description 4
- 238000012360 testing method Methods 0.000 claims description 4
- 230000000007 visual effect Effects 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 abstract description 3
- 238000011160 research Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 3
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000011273 social behavior Effects 0.000 description 2
- 238000004220 aggregation Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 239000003795 chemical substances by application Substances 0.000 description 1
- 238000005094 computer simulation Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000007591 painting process Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000013517 stratification Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Landscapes
- Document Processing Apparatus (AREA)
Abstract
The invention discloses a kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar.Method is handled based on the artificial intelligence technology of using the Chinese character shape grammar and is imitated the multiple uncertain factor that occurs in the process.At first the Chinese calligraphy word is resolved into spatial relationship between fundamental element and element, then element and spatial relationship are carried out parameter expression, obtain the probability and the confidence level of element, spatial relationship, next utilize enhancing form grammer that the Chinese calligraphy word of imitating is given a mark, utilize optimized Algorithm to optimize mark at last, up to reaching satisfaction.The advantage of this method is to have utilized the ability to express of enriching of shape grammar to represent Chinese character style, has caught the potential feature in the individual person's handwriting, and can be by computer automatic execution.
Description
Technical field
The present invention relates to computer art and aesthetics and artificial intelligence field, relate in particular to a kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar.
Background technology
Had a large amount of work to carry out the correlative study of Chinese calligraphy word, method and this research method of IEEE IntelligentSystems magazine (article title " Automatic generation of artistic Chinese calligraphy ") in 2005 announcement have relation.In that piece article, use the form grammer to decompose the Chinese calligraphy word with the method for stratification.Thereby be not captured in uncertain factor in the Chinese calligraphy word formation process but strengthen the form grammer.Nineteen ninety-five Proceedings of the International Conference on ComputerProcessing of Oriental Languages magazine (article title " Chinese glyph generation usingcharacter composition and beauty evaluation metrics ") proposes a kind of aesthetic measure of estimating writing brush word with the mode of exploring in the mode that quantizes.Article draws four kinds of rules of the process of writing at the Chinese character calligraphy wordbook based on a rule-based assessment method attractive in appearance.The score summation of four kinds of rules of this method simple computation.Proceeding of the International Joint Conference on Neural Networks magazine (article title " Fuzzy theory in hand writing learning system ") proposition blur methods in 1992 are estimated the aesthetic measure of Chinese calligraphy word.Introducing member function in this blur method catches the difference of different calligraphies and writes pattern.But the design of member function is normally finished by hand and all is write pattern all fix.By contrast, our the form grammer of the enhancing that proposes obtains by dynamic training.By the process of an instant training, can catch the individual and write uncertain factor on the pattern at the Chinese calligraphy wordbook.
Another field relevant with this method is imitating of Chinese calligraphy word.Creativity and the trial in jazz production process of Proceedings of thetwelfth national conference on Artificial intelligence magazine (article title " Simulationingcreativity in jazz performance ") in 1994 by music foundation knowledge and music memorizing analog music man.The contact that the types of ACM Transactions on Graphics magazines (article title " Learning style translation for the lines of a drawing ") in 2003 by research different drawing man in different art drawing obtain, research is interesting problem in cartoon painting process shape is imitated, and this research provides a direct inspiration for our method.Framed structure based on the multilayer agency of AI Society magazine (article title " A multi-agent a based framework for the simulaion of human and socialbehaviors during emergency evacuations ") in 2007 proposition is used for the mankind and the social behavior of simulating in the emergency escape process.
Generally speaking, we observe all computer simulation systems all is by modeling moves to destination object, modeling process can be undertaken by the method for data aggregation, also can produce model by simulation, produce in the process of model in simulation, model can provide in advance by the human expert, perhaps learns gradually by online process, and perhaps both have concurrently.
Summary of the invention
The objective of the invention is to overcome the deficiencies in the prior art, a kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar is provided.
Automatic imitative writing method for personal Chinese character handwritten font based on shape grammar may further comprise the steps:
1) use shape grammar to decompose the Chinese calligraphy word
The Chinese calligraphy word is made up of basic stroke, according to shape grammar the Chinese calligraphy word is resolved into the hierarchical structure that stroke constitutes, shape grammar comprises a series of shape generation rule, every kind of rule all points out to constitute from the bottom stroke component of Chinese calligraphy word, except the component of Chinese calligraphy word, also have the space layout between another component, space layout is used to refer to the spatial relationship between the Chinese calligraphy word component;
2) catch the uncertain factor of Chinese calligraphy word by the shape grammar of strengthening
By strengthening each shape grammar rule, realize the shape grammar system of an enhancing, the process of this enhancing comprises two kinds of Writer's strokes of contact and space layout, thereby obtain the probability that certain writer writes and the confidence level of this probability, the shape grammar that strengthens no longer is consistent to all users, but rely on the writer, create the shape grammar of reinforcement for specific writer, this process is exactly to determine an optimal set, comprises in this set that all strokes and space layout are to a certain specific Writer's probability and confidence level;
3) shape grammar of use the strengthening confidence level of determining to imitate the Chinese calligraphy word
Obtain the rule that all form the Chinese calligraphy word by the reasoning in the form grammar system, rule to each Chinese calligraphy word, the probability of recursive calculation stroke and confidence level, in order to adapt to any syntax rule, adopt neural network method to produce the probability and the confidence level of stroke, utilize all form syntax rules to train these neural networks;
4) based on the confidence level of imitating the Chinese calligraphy word, the proposition method is imitated the Chinese calligraphy word
When evaluation is carried out utilizing confidence level to imitate the Chinese calligraphy word as feedback after confidence level analyzes at a certain Writer's stroke of imitating.
Described step 1) comprises:
A) the Chinese calligraphy word is expressed by a tree structure, and the branches and leaves of tree all are basic components, and wherein the leaf of the bottom is exactly the stroke in the Chinese calligraphy word, and the Chinese calligraphy word is split into basic stroke;
B) form the hierarchical structure relation of also having living space between the basic stroke of Chinese calligraphy word, by this relation, the Chinese calligraphy word is seen as the sandwich construction of being made up of a plurality of strokes;
C) according to the hierarchical structure of the Chinese calligraphy word of step a), step b), a Chinese calligraphy word is formed with the form of recurrence, finish by multiple composing method, each forms the method for Chinese calligraphy word all corresponding to a kind of analytical approach in the form grammar system;
D) the form grammer comprises a series of form generation rules, and these rules can be expressed out in the form grammar system, and each rule all points out how to form from the low layer element element of Chinese calligraphy word;
E) except the fundamental element of forming the Chinese calligraphy word, in the form grammer, also have the word predicted composition, the word predicted composition refers to the spatial relationship of forming Chinese calligraphy word fundamental element;
F) seek all composing methods of certain specific Chinese calligraphy word, finding all composing methods is a key of successfully imitating Chinese calligraphy word algorithm;
G) in finding all processes of forming the Chinese calligraphy word, each component all is given a type attribute, this attribute provides type under the element according to Chinese tradition Chinese character typoiogical classification, in case obtain the morphotype of component, can produce the true form of Chinese calligraphy word according to corresponding form syntax rule, wherein corresponding syntax rule obtains by the reasoning of form grammar system.
Described step 2) comprising:
H) the form grammar system of Zeng Qianging forms by strengthening each form syntax rule, the process that strengthens comprises element, spatial relationship in two kinds of author's relevant informations and the form syntax rule connected, and just they are by the probability that the someone write and the confidence level of this probability;
I) different with the traditional form grammar system of speaking of previously, the shape grammar that strengthens be not consistent to all users, but rely on the writer, and this is an objective attribute target attribute in design process, the uncertain factor of successfully catching a person writing is the key that the Chinese calligraphy word is imitated in success;
J) create the shape grammar of reinforcement for specific writer, this process is exactly to determine an optimal set, comprise in this set that all strokes and space layout are to a certain specific Writer's probability and confidence level, according to a certain Writer's stroke set the Chinese calligraphy word is resolved into the basic composition element, thereby produce a series of these Writer's stroke writings, these stroke writings or derive from the writer, perhaps derive from the process that the Chinese calligraphy word decomposes, then each stroke is carried out parameter expression, make it become an one-dimensional vector, preceding two parameters of vector provide the track that stroke formed in the Chinese calligraphy word, and another parameter provides the width of stroke;
K) under the situation of given Chinese calligraphy word component, determine to be write, belong to the confidence level of probability and this probability of certain form by certain specific writer, earlier in training set, find all and this word to belong to all constituent elements of form of the same race, and adopt at step j) in the method for the parameter expression mentioned, element in the training set is carried out parameter expression, these elements are all from training set, so corresponding writer knows in advance, result behind the parameter expression is divided into two groups, write by certain writer X for one group, another group is not write by certain writer X;
L) to the input Chinese calligraphy word component according to step j) method carry out parameter expression, the vector that parameter expression obtains will fall into which the group, promptly write by certain writer X, still do not write by certain writer X;
Whether m) will distinguish stroke in training set is write by a certain specific writer, next introduce a fuzzy classification data clusters algorithm, in order to deal with problems this algorithm is made an amendment slightly: replace traditional aggregate distance with curve distance, in this fuzzy clustering algorithm, export certain stroke and belong to certain Writer's probability, ten-fold crossover algorithm in the data mining is used for producing the confidence level of probability here, adopt sample data to train then, and stay 10% data and test;
N) accuracy of method classification determining step m by the following method), repeating step m) 10 time, obtain an overall classification accuracy, if overall average classification accuracy is 100%, promptly give its highest confidence level=1, classification results is correct probably when running into new element so, otherwise, many more mistakes appear in above-mentioned assorting process, and just low more to the confidence level of the classification results of new element;
O) catch the enhancing form grammer of uncertain factor in the Chinese calligraphy word for initialization, adopt the recognizer of the Chinese calligraphy word of optimizing, determine the probability and the confidence level of spatial relationship between the element of all Chinese calligraphy words, the output of this algorithm is a fuzzy digit, and each part of numeral points out that spatial relationship belongs to the probability of certain particular space relation;
P) at all elements, according to step k), step l), step m) handle, the purpose of processing is to be write, belong to the probability and the confidence level of certain particular space relation for the spatial relationship that obtains element by certain writer, and to belong to the spatial relationship of other type space big if the spatial relationship of element belongs to the likelihood ratio of certain particular space relation, illustrate that the result is correct, when all spatial relationship through after this step, calculated population accuracy, and the result exported as confidence level.
Described step 3) comprises:
Q) key of successfully imitating the Chinese calligraphy word in the computer approach of imitating the Chinese calligraphy word is to determine confidence level that certain specific calligraphist is imitated by marking, and the form grammer by enhancing previously discussed designs this scoring part;
R) obtain the rule that all form the Chinese calligraphy word by the reasoning in the form grammar system, to each composition rule, the probability of recursive calculation stroke and confidence level;
S) in order to adopt neural network method to produce the probability and the confidence level of stroke to any syntax rule.Two kinds of neural networks are used, and a kind of is the confidence level of prediction stroke
Be called
Another kind is that (P, X Y), are called NN for the probability ρ of stroke
ρBe input to
Content comprise the confidence level of the probability of certain layout in some stroke space layouts
Belong to the confidence level of the probability of certain font with certain stroke
Be input to NN
ρContent comprise probability ρ (R, the P ' of certain layout in some stroke space layouts
1, P '
l, X, Y) and certain stroke belong to certain font probability ρ (P '
s, X, Y), wherein
Output be the confidence level that certain stroke belongs to the probability of certain font
NN
ρOutput be the probability that certain stroke belongs to certain font;
T) utilize all form syntax rules to train these neural networks, wherein the component of syntax rule and space layout are all write down by certain writer, then obtain ρ and
Value, each bar rule all provides a training example to neural network, the neural network that is adopted be classical back to Feedback Neural Network, in the neural metwork training process, be made as circulation 10000 times, and adopt the ten-fold interleaving techniques to prevent over adaptation.
Described step 4) comprises:
U) carry out after confidence level analyzes at a certain Writer's stroke of imitating when evaluation, the Chinese calligraphy word generating algorithm that adopts 2005 " Automatic generation of artistic Chinese calligraphy " to propose utilizes confidence level to imitate the Chinese calligraphy word as feedback;
V) Chinese calligraphy word generating algorithm has six parameters, provide interface to integrate and imitate scoring algorithm, these parameters are used to refer to the visual style of the Chinese calligraphy word of generation, get earlier this six parameters at random, call then and imitate scoring algorithm and come imitating confidence level marking, after producing mark, adopt deepest ascendant algorithm to make the mark maximization, for avoiding being absorbed in local minimum, repeat The above results repeatedly, arrive a satisfied threshold 0.9 up to mark, perhaps the number of times of Chong Fuing is broken through 1000 times, by imitating the Chinese calligraphy word one by one, imitate whole Chinese calligraphy words.
The beneficial effect that the present invention compared with prior art has:
(1) combine the kinds of artificial intellectual technology, making computing machine imitate the Chinese calligraphy word becomes possibility;
(2) utilize nerual network technique, make computing machine can come by the mankind's aesthetic conceptions study the ability that the Chinese calligraphy word carries out poor family;
(3) in the stroke to the Chinese calligraphy word decomposes, introduced probability and confidence level, the stroke that has improved greatly for the serious especially careless tree body of font of deformation decomposes effect.
Description of drawings
Fig. 1 implementing procedure figure of the present invention;
Fig. 2 (a) shows a kind of mode figure that decomposes the Chinese calligraphy word;
Fig. 2 (b) shows the another kind of mode figure that decomposes the Chinese calligraphy word;
Fig. 3 is presented at all form syntax rule figure that decompose among Fig. 1;
Fig. 4 is presented at the part form syntax rule of decomposing among Fig. 1, and with another kind of formal representation syntax rule figure;
Fig. 5 shows that a part is according to form syntax rule composition Chinese calligraphy word procedure chart;
First row are original writing brush word that the calligraphist writes among Fig. 6, and other three row are our algorithm produces, the highest figure as a result that imitate that imitate the confidence level score of acquisition.
Embodiment
Automatic imitative writing method for personal Chinese character handwritten font based on shape grammar may further comprise the steps:
1) use shape grammar to decompose the Chinese calligraphy word
The Chinese calligraphy word is made up of basic stroke, according to shape grammar the Chinese calligraphy word is resolved into the hierarchical structure that stroke constitutes, shape grammar comprises a series of shape generation rule, every kind of rule all points out to constitute from the bottom stroke component of Chinese calligraphy word, except the component of Chinese calligraphy word, also have the space layout between another component, space layout is used to refer to the spatial relationship between the Chinese calligraphy word component;
2) catch the uncertain factor of Chinese calligraphy word by the shape grammar of strengthening
By strengthening each shape grammar rule, realize the shape grammar system of an enhancing, the process of this enhancing comprises two kinds of Writer's strokes of contact and space layout, thereby obtain the probability that certain writer writes and the confidence level of this probability, the shape grammar that strengthens no longer is consistent to all users, but rely on the writer, create the shape grammar of reinforcement for specific writer, this process is exactly to determine an optimal set, comprises in this set that all strokes and space layout are to a certain specific Writer's probability and confidence level;
3) shape grammar of use the strengthening confidence level of determining to imitate the Chinese calligraphy word
Obtain the rule that all form the Chinese calligraphy word by the reasoning in the form grammar system, rule to each Chinese calligraphy word, the probability of recursive calculation stroke and confidence level, in order to adapt to any syntax rule, adopt neural network method to produce the probability and the confidence level of stroke, utilize all form syntax rules to train these neural networks;
4) based on the confidence level of imitating the Chinese calligraphy word, the proposition method is imitated the Chinese calligraphy word
When evaluation is carried out utilizing confidence level to imitate the Chinese calligraphy word as feedback after confidence level analyzes at a certain Writer's stroke of imitating.
Described step 1) comprises:
A) the Chinese calligraphy word is expressed by a tree structure, and the branches and leaves of tree all are basic components, and wherein the leaf of the bottom is exactly the stroke in the Chinese calligraphy word, and the Chinese calligraphy word is split into basic stroke;
B) form the hierarchical structure relation of also having living space between the basic stroke of Chinese calligraphy word, by this relation, the Chinese calligraphy word is seen as the sandwich construction of being made up of a plurality of strokes;
C) according to the hierarchical structure of the Chinese calligraphy word of step a), step b), a Chinese calligraphy word is formed with the form of recurrence, finish by multiple composing method, each forms the method for Chinese calligraphy word all corresponding to a kind of analytical approach in the form grammar system;
D) the form grammer comprises a series of form generation rules, and these rules can be expressed out in the form grammar system, and each rule all points out how to form from the low layer element element of Chinese calligraphy word;
E) except the fundamental element of forming the Chinese calligraphy word, in the form grammer, also have the word predicted composition, the word predicted composition refers to the spatial relationship of forming Chinese calligraphy word fundamental element;
F) seek all composing methods of certain specific Chinese calligraphy word, finding all composing methods is a key of successfully imitating Chinese calligraphy word algorithm;
G) in finding all processes of forming the Chinese calligraphy word, each component all is given a type attribute, this attribute provides type under the element according to Chinese tradition Chinese character typoiogical classification, in case obtain the morphotype of component, can produce the true form of Chinese calligraphy word according to corresponding form syntax rule, wherein corresponding syntax rule obtains by the reasoning of form grammar system.
Described step 2) comprising:
H) the form grammar system of Zeng Qianging forms by strengthening each form syntax rule, the process that strengthens comprises element, spatial relationship in two kinds of author's relevant informations and the form syntax rule connected, and just they are by the probability that the someone write and the confidence level of this probability;
I) different with the traditional form grammar system of speaking of previously, the shape grammar that strengthens be not consistent to all users, but rely on the writer, and this is an objective attribute target attribute in design process, the uncertain factor of successfully catching a person writing is the key that the Chinese calligraphy word is imitated in success;
J) create the shape grammar of reinforcement for specific writer, this process is exactly to determine an optimal set, comprise in this set that all strokes and space layout are to a certain specific Writer's probability and confidence level, according to a certain Writer's stroke set the Chinese calligraphy word is resolved into the basic composition element, thereby produce a series of these Writer's stroke writings, these stroke writings or derive from the writer, perhaps derive from the process that the Chinese calligraphy word decomposes, then each stroke is carried out parameter expression, make it become an one-dimensional vector, preceding two parameters of vector provide the track that stroke formed in the Chinese calligraphy word, and another parameter provides the width of stroke;
K) under the situation of given Chinese calligraphy word component, determine to be write, belong to the confidence level of probability and this probability of certain form by certain specific writer, earlier in training set, find all and this word to belong to all constituent elements of form of the same race, and adopt at step j) in the method for the parameter expression mentioned, element in the training set is carried out parameter expression, these elements are all from training set, so corresponding writer knows in advance, result behind the parameter expression is divided into two groups, write by certain writer X for one group, another group is not write by certain writer X;
L) to the input Chinese calligraphy word component according to step j) method carry out parameter expression, the vector that parameter expression obtains will fall into which the group, promptly write by certain writer X, still do not write by certain writer X;
Whether m) will distinguish stroke in training set is write by a certain specific writer, next introduce a fuzzy classification data clusters algorithm, in order to deal with problems this algorithm is made an amendment slightly: replace traditional aggregate distance with curve distance, in this fuzzy clustering algorithm, export certain stroke and belong to certain Writer's probability, ten-fold crossover algorithm in the data mining is used for producing the confidence level of probability here, adopt sample data to train then, and stay 10% data and test;
N) accuracy of method classification determining step m by the following method), repeating step m) 10 time, obtain an overall classification accuracy, if overall average classification accuracy is 100%, promptly give its highest confidence level=1, classification results is correct probably when running into new element so, otherwise, many more mistakes appear in above-mentioned assorting process, and just low more to the confidence level of the classification results of new element;
O) catch the enhancing form grammer of uncertain factor in the Chinese calligraphy word for initialization, adopt the recognizer of the Chinese calligraphy word of optimizing, determine the probability and the confidence level of spatial relationship between the element of all Chinese calligraphy words, the output of this algorithm is a fuzzy digit, and each part of numeral points out that spatial relationship belongs to the probability of certain particular space relation;
P) at all elements, according to step k), step l), step m) handle, the purpose of processing is to be write, belong to the probability and the confidence level of certain particular space relation for the spatial relationship that obtains element by certain writer, and to belong to the spatial relationship of other type space big if the spatial relationship of element belongs to the likelihood ratio of certain particular space relation, illustrate that the result is correct, when all spatial relationship through after this step, calculated population accuracy, and the result exported as confidence level.
Described step 3) comprises:
Q) key of successfully imitating the Chinese calligraphy word in the computer approach of imitating the Chinese calligraphy word is to determine confidence level that certain specific calligraphist is imitated by marking, and the form grammer by enhancing previously discussed designs this scoring part;
R) obtain the rule that all form the Chinese calligraphy word by the reasoning in the form grammar system, to each composition rule, the probability of recursive calculation stroke and confidence level;
S) in order to adopt neural network method to produce the probability and the confidence level of stroke to any syntax rule.Two kinds of neural networks are used, and a kind of is the confidence level of prediction stroke
Be called
Another kind is that (P, X Y), are called NN for the probability ρ of stroke
ρBe input to
Content comprise the confidence level of the probability of certain layout in some stroke space layouts
Belong to the confidence level of the probability of certain font with certain stroke
Be input to NN
ρContent comprise probability ρ (R, the P ' of certain layout in some stroke space layouts
1... P '
l, X, Y) and certain stroke belong to certain font probability ρ (P '
s, X, Y), wherein
Output be the confidence level that certain stroke belongs to the probability of certain font
Output be the probability that certain stroke belongs to certain font;
T) utilize all form syntax rules to train these neural networks, wherein the component of syntax rule and space layout are all write down by certain writer, then obtain ρ and
Value, each bar rule all provides a training example to neural network, the neural network that is adopted be classical back to Feedback Neural Network, in the neural metwork training process, be made as circulation 10000 times, and adopt the ten-fold interleaving techniques to prevent over adaptation.
Described step 4) comprises:
U) carry out after confidence level analyzes at a certain Writer's stroke of imitating when evaluation, the Chinese calligraphy word generating algorithm that adopts 2005 " Automatic generation of artistic Chinese calligraphy " to propose utilizes confidence level to imitate the Chinese calligraphy word as feedback;
V) Chinese calligraphy word generating algorithm has six parameters, provide interface to integrate and imitate scoring algorithm, these parameters are used to refer to the visual style of the Chinese calligraphy word of generation, get earlier this six parameters at random, call then and imitate scoring algorithm and come imitating confidence level marking, after producing mark, adopt deepest ascendant algorithm to make the mark maximization, for avoiding being absorbed in local minimum, repeat The above results repeatedly, arrive a satisfied threshold 0.9 up to mark, perhaps the number of times of Chong Fuing is broken through 1000 times, by imitating the Chinese calligraphy word one by one, imitate whole Chinese calligraphy words.
Embodiment
As shown in Figure 1, the flow process of implementation system of the present invention comprises Chinese calligraphy word image 101, spatial relationship decomposes 102 between stroke and stroke, utilizes to strengthen the uncertain factor 103 that shape grammar is caught the Chinese calligraphy word, utilizes the evaluation of enhancing shape grammar to imitate Chinese calligraphy word 104;
Chinese calligraphy word image 101: the Chinese calligraphy word image is meant the digital picture that comprises Chinese character style; In the present embodiment, all Chinese calligraphy word images all have been separated into individual character one by one, then they are normalized into the two-value black white image of unified size, and the example is shown in Fig. 6 first row;
Spatial relationship decomposes 102 between stroke and stroke: in this example, this part may further comprise the steps:
A) the Chinese calligraphy word can be expressed by a tree structure.The branches and leaves of tree all are basic components, and wherein the leaf of the bottom is exactly the stroke in the Chinese calligraphy word.Therefore the Chinese calligraphy word can be split into basic stroke.Form the hierarchical structure relation of also having living space between the basic stroke of Chinese calligraphy word, by this relation, the Chinese calligraphy word can be seen as the sandwich construction of being made up of a plurality of strokes.
B) pass through A) in the hierarchical structure of the Chinese calligraphy word discussed, a Chinese calligraphy word can be formed by the form of recurrence.This forming process can be finished in several ways.Each forms the method for Chinese calligraphy word all corresponding to a kind of analytical approach in the form grammar system.Why Here it is selects the form grammer as the basic production method that produces the Chinese calligraphy word.
C) the form grammer comprises a series of form generation rules.These rules can be expressed out in the form grammar system.Each rule all points out how to form from the low layer element element of Chinese calligraphy word.Except the fundamental element of forming the Chinese calligraphy word, the word predicted composition is also very important in the form grammer.The word predicted composition refers to the spatial relationship of forming Chinese calligraphy word fundamental element.
D) use above step, we can list all possible mode of forming certain specific Chinese calligraphy word.Find key of successfully imitating Chinese calligraphy word algorithm of all composition modes.In finding all processes of forming the Chinese calligraphy word, each component all is given a type attribute.This attribute provides type under the element according to Chinese tradition Chinese character typoiogical classification.In case obtain the morphotype of component, we can produce the true form of Chinese calligraphy word according to corresponding form syntax rule, and wherein corresponding syntax rule obtains by the reasoning of form grammar system.Provide whole decomposable process and tree structure utilization among Fig. 2 and strengthen the uncertain factor 103 that the Chinese calligraphy word caught in the form grammer:
E) the form grammar system of Zeng Qianging forms by strengthening each form syntax rule.The process that strengthens comprises element, spatial relationship in two kinds of author's relevant informations and the form syntax rule is connected.Just they are by the confidence level of probability that the someone write and this probability.Its step is as follows:
1) ρ (P
i, X, Y): Chinese calligraphy Character table Y belongs to certain form τ (P
i) and the probability write by writer X;
2)
Confidence level;
3) ρ (left_notouch, P
1, P
2, X, Y): element P
1And P
2Between the spatial relationship Y probability that belongs to certain type τ (left_notouch) and write by writer X;
4)
Confidence level.
F) different with the traditional form grammar system of speaking of previously, the shape grammar of enhancing be not consistent to all users, but relies on the writer.This is an objective attribute target attribute in design process.The uncertain factor of successfully catching a person writing is the key that the Chinese calligraphy word is imitated in success.
G) create the shape grammar of reinforcement for specific writer.This process is exactly to determine an optimal set, comprises in this set that all strokes and space layout are to a certain specific Writer's probability and confidence level.Can resolve into the basic composition element to the Chinese calligraphy word according to a certain Writer's stroke set, thereby produce a series of these Writer's stroke writings.These stroke writings or derive from the writer perhaps derive from the process that the Chinese calligraphy word decomposes, and then each stroke are carried out parameter expression, make it become an one-dimensional vector.Preceding two parameters of vector provide the track that stroke formed in the Chinese calligraphy word, and another parameter provides the width of stroke.
H) under the situation of given Chinese calligraphy word component, determine to be write, belong to the confidence level of probability and this probability of certain form by certain specific writer.Earlier in training set, find all and this word to belong to all constituent elements of form of the same race.And adopt at j) in the method for the parameter expression mentioned, the element in the training set is carried out parameter expression.Because these elements all from training set, are not known in advance so write the person accordingly.We can be divided into the result behind the parameter expression two groups, are write by certain writer X for one group, and another group is not write by certain writer X.
I) we to the Chinese calligraphy word component of input according to j) method carry out parameter expression, now problem just is which group the vector that parameter expression obtains will fall into, and promptly be write by certain writer X, still not write by certain writer X.
J) in training set, to distinguish stroke and whether be write, next introduce a fuzzy classification data clusters algorithm, this algorithm be made an amendment slightly: replace traditional aggregate distance with curve distance in order to deal with problems by a certain specific writer.In this fuzzy clustering algorithm, can export certain stroke and belong to certain Writer's probability.Ten-fold crossover algorithm in the data mining is used for producing the confidence level of probability here.We adopt sample data to train then, and stay 10% data and test.
K) we determine m by the following method) in the accuracy of method classification.We repeat said process 10 times, obtain an overall classification accuracy.If overall average classification accuracy is 100%, promptly we give its highest confidence level (=1), and classification results is correct probably when running into new element so.Otherwise, many more mistakes appear in above-mentioned assorting process, and just low more to the confidence level of the classification results of new element.
L) catch the enhancing form grammer of uncertain factor in the Chinese calligraphy word for initialization, we need determine the probability and the confidence level of spatial relationship between the element of all Chinese calligraphy words.A lot of researchs have been carried out about the identification of optimizing the Chinese calligraphy word.We adopt a kind of easy realization and effective algorithm.The output of this algorithm is a fuzzy digit, and each part of numeral points out that spatial relationship belongs to the probability of certain particular space relation.
M) next according to k), l), the step of mentioning in m) is similarly handled.The purpose of handling is to be write, belong to the probability and the confidence level of certain particular space relation by certain writer for the spatial relationship that obtains element.To belong to other type space spatial relationship big if the spatial relationship of element belongs to the likelihood ratio of certain particular space relation, and we just we can say that the result is correct.When all spatial relationship through after this step, the correct number percent of our calculated population, and the result exported as confidence level.
Mark 104 to imitating the Chinese calligraphy word:
N) key of successfully imitating the Chinese calligraphy word in the computer approach of imitating the Chinese calligraphy word is to determine confidence level that certain specific calligraphist is imitated by marking.Form grammer by enhancing previously discussed designs this scoring part.
O) obtain the rule that all form the Chinese calligraphy word by the reasoning in the form grammar system.To each composition rule, the probability of recursive calculation stroke and confidence level.Detailed step is as follows:
1) supposes that the strictly all rules result is W={W
1, W
2..., W
n, to each regular recursive calculation ρ (P, X, Y) and
Make W
iLength be m.
2) if m=1, W
iJust only contain a syntax rule, as shown in Figure 4.
3) if m＞1, we are with regard to the equation of each form grammer of repeated application.In this repetitive process, can obtain probability and confidence level that element Y belongs to certain font τ (P) and write by writer X.
4) the following function of definition is considered probability and two factors of confidence level:
Wherein k is a parameter that the user can regulate.
5) last in the result who produces by all composition rules, we are selective rule W
i, this rule can maximize arg max
iO
i(P, X, Y)
P) in order to adopt neural network method to produce the probability and the confidence level of stroke to any syntax rule.Two kinds of neural networks are used, and a kind of is the confidence level of prediction stroke
Be called
Another kind is that (P, X Y), are called NN for the probability ρ of stroke
ρBe input to
Content comprise the confidence level of the probability of certain layout in some stroke space layouts
Belong to the confidence level of the probability of certain font with certain stroke
Be input to NN
ρContent comprise probability ρ (R, the P ' of certain layout in some stroke space layouts
1' ..., P '
l, X, Y) and certain stroke belong to certain font probability ρ (P '
s, X, Y).Wherein
Output be the confidence level that certain stroke belongs to the probability of certain font
NN
ρOutput be the probability that certain stroke belongs to certain font.
Q) utilize all form syntax rules to train these neural networks, wherein the component of syntax rule and space layout are all write down by certain writer, thus just mean ρ and
Value all known.Each bar rule all provides a training example to neural network.Here the neural network of Cai Yonging is that classical back is to Feedback Neural Network.In the neural metwork training process, we are made as circulation 10000 times, and adopt the ten-fold interleaving techniques in order to prevent over adaptation.
Imitate Chinese calligraphy word 105:
U) carry out just can utilizing confidence level to imitate the Chinese calligraphy word after confidence level analyzes at a certain Writer's stroke of imitating when our evaluation as feeding back.Here, the Chinese calligraphy word generating algorithm that adopts 2005 " Automaticgeneration of artistic Chinese calligraphy " to propose.
V) this algorithm has six parameters, provides a good interface to integrate and has imitated scoring algorithm.These parameters are used to refer to the visual style of the Chinese calligraphy word that is burdened with.We earlier get this six parameters at random, call then to imitate scoring algorithm and come imitating confidence level marking.After producing mark, we adopt the deepestascendant algorithm to make the mark maximization.For avoiding being absorbed in local minimum, we repeat The above results repeatedly, know that mark arrives a satisfied threshold (0.9), and perhaps the number of times of Chong Fuing is broken through 1000 times.By imitating the Chinese calligraphy word one by one, we can imitate whole Chinese calligraphy words.
Claims (5)
1. automatic imitative writing method for personal Chinese character handwritten font based on shape grammar is characterized in that may further comprise the steps:
1) use shape grammar to decompose the Chinese calligraphy word
The Chinese calligraphy word is made up of basic stroke, according to shape grammar the Chinese calligraphy word is resolved into the hierarchical structure that stroke constitutes, shape grammar comprises a series of shape generation rule, every kind of rule all points out to constitute from the bottom stroke component of Chinese calligraphy word, except the component of Chinese calligraphy word, also have the space layout between another component, space layout is used to refer to the spatial relationship between the Chinese calligraphy word component;
2) catch the uncertain factor of Chinese calligraphy word by the shape grammar of strengthening
By strengthening each shape grammar rule, realize the shape grammar system of an enhancing, the process of this enhancing comprises two kinds of Writer's strokes of contact and space layout, thereby obtain the probability that certain writer writes and the confidence level of this probability, the shape grammar that strengthens no longer is consistent to all users, but rely on the writer, create the shape grammar of reinforcement for specific writer, this process is exactly to determine an optimal set, comprises in this set that all strokes and space layout are to a certain specific Writer's probability and confidence level;
3) shape grammar of use the strengthening confidence level of determining to imitate the Chinese calligraphy word
Obtain the rule that all form the Chinese calligraphy word by the reasoning in the form grammar system, rule to each Chinese calligraphy word, the probability of recursive calculation stroke and confidence level, in order to adapt to any syntax rule, adopt neural network method to produce the probability and the confidence level of stroke, utilize all form syntax rules to train these neural networks;
4) based on the confidence level of imitating the Chinese calligraphy word, the proposition method is imitated the Chinese calligraphy word
When evaluation is carried out utilizing confidence level to imitate the Chinese calligraphy word as feedback after confidence level analyzes at a certain Writer's stroke of imitating.
2. a kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar according to claim 1 is characterized in that described step 1) comprises:
A) the Chinese calligraphy word is expressed by a tree structure, and the branches and leaves of tree all are basic components, and wherein the leaf of the bottom is exactly the stroke in the Chinese calligraphy word, and the Chinese calligraphy word is split into basic stroke;
B) form the hierarchical structure relation of also having living space between the basic stroke of Chinese calligraphy word, by this relation, the Chinese calligraphy word is seen as the sandwich construction of being made up of a plurality of strokes;
C) according to the hierarchical structure of the Chinese calligraphy word of step a), step b), a Chinese calligraphy word is formed with the form of recurrence, finish by multiple composing method, each forms the method for Chinese calligraphy word all corresponding to a kind of analytical approach in the form grammar system;
D) the form grammer comprises a series of form generation rules, and these rules can be expressed out in the form grammar system, and each rule all points out how to form from the low layer element element of Chinese calligraphy word;
E) except the fundamental element of forming the Chinese calligraphy word, in the form grammer, also have the word predicted composition, the word predicted composition refers to the spatial relationship of forming Chinese calligraphy word fundamental element;
F) seek all composing methods of certain specific Chinese calligraphy word, finding all composing methods is a key of successfully imitating Chinese calligraphy word algorithm;
G) in finding all processes of forming the Chinese calligraphy word, each component all is given a type attribute, this attribute provides type under the element according to Chinese tradition Chinese character typoiogical classification, in case obtain the morphotype of component, can produce the true form of Chinese calligraphy word according to corresponding form syntax rule, wherein corresponding syntax rule obtains by the reasoning of form grammar system.
3. the automatic imitative writing method for personal Chinese character handwritten font based on shape grammar according to claim 1 is characterized in that described step 2) comprising:
H) the form grammar system of Zeng Qianging forms by strengthening each form syntax rule, the process that strengthens comprises element, spatial relationship in two kinds of author's relevant informations and the form syntax rule connected, and just they are by the probability that the someone write and the confidence level of this probability;
I) different with the traditional form grammar system of speaking of previously, the shape grammar that strengthens be not consistent to all users, but rely on the writer, and this is an objective attribute target attribute in design process, the uncertain factor of successfully catching a person writing is the key that the Chinese calligraphy word is imitated in success;
J) create the shape grammar of reinforcement for specific writer, this process is exactly to determine an optimal set, comprise in this set that all strokes and space layout are to a certain specific Writer's probability and confidence level, according to a certain Writer's stroke set the Chinese calligraphy word is resolved into the basic composition element, thereby produce a series of these Writer's stroke writings, these stroke writings or derive from the writer, perhaps derive from the process that the Chinese calligraphy word decomposes, then each stroke is carried out parameter expression, make it become an one-dimensional vector, preceding two parameters of vector provide the track that stroke formed in the Chinese calligraphy word, and another parameter provides the width of stroke;
K) under the situation of given Chinese calligraphy word component, determine to be write, belong to the confidence level of probability and this probability of certain form by certain specific writer, earlier in training set, find all and this word to belong to all constituent elements of form of the same race, and adopt at step j) in the method for the parameter expression mentioned, element in the training set is carried out parameter expression, these elements are all from training set, so corresponding writer knows in advance, result behind the parameter expression is divided into two groups, write by certain writer X for one group, another group is not write by certain writer X;
L) to the input Chinese calligraphy word component according to step j) method carry out parameter expression, the vector that parameter expression obtains will fall into which the group, promptly write by certain writer X, still do not write by certain writer X;
Whether m) will distinguish stroke in training set is write by a certain specific writer, next introduce a fuzzy classification data clusters algorithm, in order to deal with problems this algorithm is made an amendment slightly: replace traditional aggregate distance with curve distance, in this fuzzy clustering algorithm, export certain stroke and belong to certain Writer's probability, ten-fold crossover algorithm in the data mining is used for producing the confidence level of probability here, adopt sample data to train then, and stay 10% data and test;
N) accuracy of method classification determining step m by the following method), repeating step m) 10 time, obtain an overall classification accuracy, if overall average classification accuracy is 100%, promptly give its highest confidence level=1, classification results is correct probably when running into new element so, otherwise, many more mistakes appear in above-mentioned assorting process, and just low more to the confidence level of the classification results of new element;
O) catch the enhancing form grammer of uncertain factor in the Chinese calligraphy word for initialization, adopt the recognizer of the Chinese calligraphy word of optimizing, determine the probability and the confidence level of spatial relationship between the element of all Chinese calligraphy words, the output of this algorithm is a fuzzy digit, and each part of numeral points out that spatial relationship belongs to the probability of certain particular space relation;
P) at all elements, according to step k), step l), step m) handle, the purpose of processing is to be write, belong to the probability and the confidence level of certain particular space relation for the spatial relationship that obtains element by certain writer, and to belong to the spatial relationship of other type space big if the spatial relationship of element belongs to the likelihood ratio of certain particular space relation, illustrate that the result is correct, when all spatial relationship through after this step, calculated population accuracy, and the result exported as confidence level.
4. a kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar according to claim 1 is characterized in that described step 3) comprises:
Q) key of successfully imitating the Chinese calligraphy word in the computer approach of imitating the Chinese calligraphy word is to determine confidence level that certain specific calligraphist is imitated by marking, and the form grammer by enhancing previously discussed designs this scoring part;
R) obtain the rule that all form the Chinese calligraphy word by the reasoning in the form grammar system, to each composition rule, the probability of recursive calculation stroke and confidence level;
S) in order to adopt neural network method to produce the probability and the confidence level of stroke to any syntax rule.Two kinds of neural networks are used, and a kind of is the confidence level of prediction stroke
Be called
Another kind is that (P, X Y), are called NN for the probability ρ of stroke
ρBe input to
Content comprise the confidence level of the probability of certain layout in some stroke space layouts
Belong to the confidence level of the probability of certain font with certain stroke
Be input to NN
ρContent comprise probability ρ (R, the P ' of certain layout in some stroke space layouts
1..., P '
l, X, Y) and certain stroke belong to certain font probability ρ (P '
s, X, Y), wherein
Output be the confidence level that certain stroke belongs to the probability of certain font
NN
ρOutput be the probability that certain stroke belongs to certain font;
T) utilize all form syntax rules to train these neural networks, wherein the component of syntax rule and space layout are all write down by certain writer, then obtain ρ and
Value, each bar rule all provides a training example to neural network, the neural network that is adopted be classical back to Feedback Neural Network, in the neural metwork training process, be made as circulation 10000 times, and adopt the ten-fold interleaving techniques to prevent over adaptation.
5. a kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar according to claim 1 is characterized in that described step 4) comprises:
U) carry out after confidence level analyzes at a certain Writer's stroke of imitating when evaluation, the Chinese calligraphy word generating algorithm that adopts 2005 " Automatic generation of artistic Chinese calligraphy " to propose utilizes confidence level to imitate the Chinese calligraphy word as feedback;
V) Chinese calligraphy word generating algorithm has six parameters, provide interface to integrate and imitate scoring algorithm, these parameters are used to refer to the visual style of the Chinese calligraphy word of generation, get earlier this six parameters at random, call then and imitate scoring algorithm and come imitating confidence level marking, after producing mark, adopt deepest ascendant algorithm to make the mark maximization, for avoiding being absorbed in local minimum, repeat The above results repeatedly, arrive a satisfied threshold 0.9 up to mark, perhaps the number of times of Chong Fuing is broken through 1000 times, by imitating the Chinese calligraphy word one by one, imitate whole Chinese calligraphy words.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2009101006502A CN101604451A (en) | 2009-07-13 | 2009-07-13 | A kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNA2009101006502A CN101604451A (en) | 2009-07-13 | 2009-07-13 | A kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN101604451A true CN101604451A (en) | 2009-12-16 |
Family
ID=41470168
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNA2009101006502A Pending CN101604451A (en) | 2009-07-13 | 2009-07-13 | A kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101604451A (en) |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101895626A (en) * | 2010-05-19 | 2010-11-24 | 济南北秀信息技术有限公司 | Handwriting analyzing device and method thereof for mobile phone |
| CN102306308A (en) * | 2011-08-26 | 2012-01-04 | 厦门大学 | Electronic brush modeling method based on texture learning |
| CN102346664A (en) * | 2011-07-22 | 2012-02-08 | 镇江诺尼基智能技术有限公司 | Automatic component layer analysis method for Chinese characters |
| CN103914503A (en) * | 2012-12-28 | 2014-07-09 | 叶青湖 | System and method for generating personalized handwriting font |
| CN106611172A (en) * | 2015-10-23 | 2017-05-03 | 北京大学 | Style learning-based Chinese character synthesis method |
| CN108804397A (en) * | 2018-06-12 | 2018-11-13 | 华南理工大学 | A method of the Chinese character style conversion based on a small amount of target font generates |
| WO2019169647A1 (en) * | 2018-03-05 | 2019-09-12 | Hong Kong Applied Science and Technology Research Institute Company Limited | Machine learning artificial character generation |
-
2009
- 2009-07-13 CN CNA2009101006502A patent/CN101604451A/en active Pending
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101895626A (en) * | 2010-05-19 | 2010-11-24 | 济南北秀信息技术有限公司 | Handwriting analyzing device and method thereof for mobile phone |
| CN102346664A (en) * | 2011-07-22 | 2012-02-08 | 镇江诺尼基智能技术有限公司 | Automatic component layer analysis method for Chinese characters |
| CN102346664B (en) * | 2011-07-22 | 2014-06-25 | 镇江诺尼基智能技术有限公司 | Automatic component layer analysis method for Chinese characters |
| CN102306308A (en) * | 2011-08-26 | 2012-01-04 | 厦门大学 | Electronic brush modeling method based on texture learning |
| CN102306308B (en) * | 2011-08-26 | 2013-07-03 | 厦门大学 | Electronic brush modeling method based on texture learning |
| CN103914503A (en) * | 2012-12-28 | 2014-07-09 | 叶青湖 | System and method for generating personalized handwriting font |
| CN106611172A (en) * | 2015-10-23 | 2017-05-03 | 北京大学 | Style learning-based Chinese character synthesis method |
| CN106611172B (en) * | 2015-10-23 | 2019-11-08 | 北京大学 | A kind of Chinese character synthetic method based on style study |
| WO2019169647A1 (en) * | 2018-03-05 | 2019-09-12 | Hong Kong Applied Science and Technology Research Institute Company Limited | Machine learning artificial character generation |
| CN108804397A (en) * | 2018-06-12 | 2018-11-13 | 华南理工大学 | A method of the Chinese character style conversion based on a small amount of target font generates |
| CN108804397B (en) * | 2018-06-12 | 2021-07-20 | 华南理工大学 | Chinese character font conversion generation method based on small amount of target fonts |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110378985B (en) | Animation drawing auxiliary creation method based on GAN | |
| CN106384094B (en) | A kind of Chinese word library automatic generation method based on writing style modeling | |
| CN101604451A (en) | A kind of automatic imitative writing method for personal Chinese character handwritten font based on shape grammar | |
| CN107066445B (en) | The deep learning method of one attribute emotion word vector | |
| CN106503654A (en) | A kind of face emotion identification method based on the sparse autoencoder network of depth | |
| CN101377854B (en) | Method for simulating Chinese characters hand-written handwriting by a computer | |
| CN108875807A (en) | A kind of Image Description Methods multiple dimensioned based on more attentions | |
| Lian et al. | EasyFont: a style learning-based system to easily build your large-scale handwriting fonts | |
| CN112199608B (en) | Social media rumor detection method based on network information propagation graph modeling | |
| CN108304826A (en) | Facial expression recognizing method based on convolutional neural networks | |
| CN107423442A (en) | Method and system, storage medium and computer equipment are recommended in application based on user's portrait behavioural analysis | |
| CN108595601A (en) | A kind of long text sentiment analysis method incorporating Attention mechanism | |
| CN106845499A (en) | A kind of image object detection method semantic based on natural language | |
| CN108197294A (en) | A kind of text automatic generation method based on deep learning | |
| CN109086270B (en) | Automatic poetry making system and method based on ancient poetry corpus vectorization | |
| CN109977861A (en) | Offline handwritten form method for identifying mathematical formula | |
| CN110334724B (en) | Remote sensing object natural language description and multi-scale correction method based on LSTM | |
| CN110415071B (en) | Automobile competitive product comparison method based on viewpoint mining analysis | |
| CN101599180B (en) | Automatic generation method of imitative computer calligraphy based on handwriting style | |
| Lake | Towards more human-like concept learning in machines: Compositionality, causality, and learning-to-learn | |
| CN111814453A (en) | Fine-grained emotion analysis method based on BiLSTM-TextCNN | |
| CN111104975A (en) | Credit assessment model based on breadth learning | |
| CN110309510B (en) | C-S and GRU-based painting and calligraphy observation method | |
| CN108519976A (en) | The method for generating extensive sentiment dictionary based on neural network | |
| Hayashi et al. | A study of data augmentation for handwritten character recognition using deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C02 | Deemed withdrawal of patent application after publication (patent law 2001) | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20091216 |