CN110570481A - calligraphy word stock automatic repairing method and system based on style migration - Google Patents
calligraphy word stock automatic repairing method and system based on style migration Download PDFInfo
- Publication number
- CN110570481A CN110570481A CN201910702258.9A CN201910702258A CN110570481A CN 110570481 A CN110570481 A CN 110570481A CN 201910702258 A CN201910702258 A CN 201910702258A CN 110570481 A CN110570481 A CN 110570481A
- Authority
- CN
- China
- Prior art keywords
- font
- generator
- image
- discriminator
- style
- 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 48
- 230000005012 migration Effects 0.000 title claims abstract description 19
- 238000013508 migration Methods 0.000 title claims abstract description 19
- 238000006243 chemical reaction Methods 0.000 claims abstract description 6
- 238000012549 training Methods 0.000 claims description 13
- 238000012545 processing Methods 0.000 claims description 8
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 5
- 230000006870 function Effects 0.000 abstract description 30
- 230000008569 process Effects 0.000 abstract description 11
- 238000007781 pre-processing Methods 0.000 description 7
- 238000013459 approach Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000000354 decomposition reaction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000000605 extraction Methods 0.000 description 3
- 230000003042 antagnostic effect Effects 0.000 description 2
- 230000008034 disappearance Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000000137 annealing Methods 0.000 description 1
- 230000008485 antagonism Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001427 coherent effect Effects 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012797 qualification Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06T11/00—2D [Two Dimensional] image generation
- G06T11/001—Texturing; Colouring; Generation of texture or colour
Abstract
the invention provides a method and a system for automatically repairing a calligraphy font library based on style migration, which comprises the following steps: setting an input font and a standard style font; inputting the input font image into an encoding module, wherein the encoding module obtains potential characteristic information; the conversion module converts the character information of the standard style font; the decoding module processes the character image to obtain a generated character image; inputting the input font image and the generated font image into a discriminator, and outputting the probability that the generated font image is a real standard style font; similarly, inputting the input font image and the standard style font image into a discriminator to obtain the probability that the standard style font image is a real standard style font; finally, obtaining loss functions of the generator and the discriminator according to the two probabilities; the optimizer adjusts the generator and the discriminator according to the loss function until the loss functions of the generator and the discriminator are converged to obtain a generator which is trained; and a complete word stock of standard style fonts can be obtained by adopting the trained generator.
Description
Technical Field
The invention relates to the technical field of character restoration, in particular to a calligraphy character library automatic restoration method and system based on style migration.
background
The calligraphy is the main expression form of Chinese character culture, not only is regarded as the artistic expression form of language, but also is used as a self-expression and culture means. Calligraphy characters can bring pleasant visual experience to people, and the working efficiency of users is improved. However, most of the calligraphy fonts which are streamed to the present world contain a very small number of Chinese characters, and a complete commercial calligraphy character library cannot be formed. For example, the regular script "mysterious tower monument" of Liugong right has a character 1512, the regular script "multi-pagoda monument" of Yan Zhen Qing N has a character 2244, and the "Tian Xian xing Shu" has a character 324. Wherein, the repeated characters of more than 820 Chinese characters are left in the 'more pagoda tablet', the number of characters required by a complete commercial word stock on the market of China is 6763 Chinese characters, GB18030 used by the official part is required to reach 27533 Chinese characters, and the number of incomplete calligraphy characters is far from meeting the daily calligraphy learning, using and researching of people. However, manually restoring a complete set of calligraphy fonts usually requires a lot of time and labor, and the repairing precision cannot be guaranteed.
Similar published patents exist:
A inscription restoration method based on Chinese character image contour feature description comprises the following steps: G06T5/40, classification number: G06T 5/40. The invention starts from the computer virtual reality technology, machine vision and image processing fields, and limits the information used for repairing the inscription to the same inscription or the calligraphy works of the same author. In the preprocessing process, the single words of the inscription are segmented, and the extracted single words are subjected to component stroke decomposition. And combining the acquired Chinese character components and strokes into a component stroke template set. In the process of repairing the damaged Chinese character, firstly, the structure information and the outline information of the credible part or stroke of the damaged Chinese character are extracted, and the template with the highest similarity is searched from the constructed template set by utilizing the structure matching and the partial correspondence of the outline segment to be used as an information source for repairing the damaged part for repairing. Compared with other image restoration methods, the method can restore the structural information and the outline detail information of the defective part of the Chinese character on the basis of better keeping the original qualification of the inscription. And the rest Chinese characters in the character library can be recombined by utilizing the decomposed strokes to design a set of complete calligraphy character library. However, this method requires the decomposition of the strokes of the calligraphy, does not handle the calligraphy with continuous strokes, such as line writing, and also takes a lot of effort to decompose the strokes.
A robot calligraphy writing method based on a generative confrontation network comprises the following steps: G06K9/00, classification number: G06K 9/00; G06K 9/62; G06N 3/04. The invention provides a generating mechanism capable of automatically generating strokes of various styles, and solves the difficulty that the existing calligraphy robot consumes a large amount of manpower and is manually input. The method specifically comprises the following steps: collecting stroke data of standard Chinese brush words, sorting according to stroke types, and marking; training two deep neural networks based on the generated confrontation network model to generate a network G and a confrontation network D; inputting the randomly sampled vector into a generation network G to obtain the probability distribution of the stroke track points; step four, the calligraphy robot acquires stroke position information from the probability distribution by applying a sampling method, writes a stroke drawing board, and shoots and records a stroke image by a camera after writing; and step five, preprocessing the image to be processed, inputting the image to the confrontation network D in the step two for training, and adjusting parameters to achieve convergence. This method, although capable of being automatically generated using a computer, has two problems: 1. the vector sampled randomly is used as input, and specified calligraphy Chinese characters cannot be generated; 2. the image needs to be preprocessed, and the standard brush-calligraphy stroke data needs to be collected and labeled, which is very troublesome.
A Chinese character font generation method for generating an countermeasure network based on conditions comprises the following steps: G06T11/00, classification number: G06T 11/00; G06T 7/13; G06F 17/21. The invention discloses a Chinese character font generation method for generating an confrontation network based on conditions, and relates to Chinese character font generation. Extracting stroke information of the Chinese character bitmap by using a central skeleton extraction and coherent point drift method; generating new style strokes with conditional antagonism generation network. The method for extracting strokes is widened, and stroke extraction is expanded from handwriting fonts without thickness information to Chinese character bitmaps with stroke thickness information. Compared with the above patent, the method can generate specific calligraphy Chinese characters based on the condition generation countermeasure network, but complex image preprocessing of extracting stroke information of Chinese character bitmaps is also needed, and the efficiency is low.
Disclosure of Invention
The invention aims to solve the technical problems that the existing calligraphy font repairing technology is low in efficiency and single in processing method, and provides a calligraphy font library automatic repairing method and system based on style migration to solve the technical defects.
The calligraphy font library automatic repairing method based on style migration comprises a generator, a discriminator, an optimizer, an encoding module, a decoding module and a converting module, wherein the generator comprises a first module, a second module and a third module; the method comprises the following specific steps:
S1, setting a printing font of the existing complete font library as an input font, and setting a calligraphy font to be repaired as a standard style font;
S2, taking the input font image as the input of a coding module in the generator, and carrying out convolution, batch standardization and activation function processing on the input font image by the coding module to obtain the potential characteristic information of the image; the conversion module converts the potential characteristic information into characteristic information of a standard style font by using the denseblock; finally, the decoding module carries out deconvolution, batch standardization and activation function processing on the converted characteristic information of the standard style font to obtain a generated font image;
S3, taking the input font image and the generated font image as the input of a discriminator, extracting font characteristic information through 5 layers of convolution of the discriminator, and finally outputting the probability that the generated font image is a real standard style font through three layers of full connection; in the same way, the input font image and the standard style font image are used as the input of the discriminator to obtain the probability that the standard style font image is the real standard style font; finally, obtaining a loss function of the generator and a loss function of the discriminator according to the two probabilities;
S4, the optimizer adjusts the network weights of the generator and the discriminator according to the loss function of the generator and the loss function of the discriminator, and repeatedly executes the steps S2-S4 after adjustment until the loss function of the generator and the loss function of the discriminator are converged to obtain a generator which is trained;
And S5, inputting all Chinese character images in the character library of the input fonts into the generator after training, generating complete Chinese character images with standard style fonts after the generator processing, and obtaining the complete character library with the standard style fonts after the Chinese character images with the standard style fonts are integrated and processed.
Further, a Wasserstein distance is introduced in the loss function.
further, the optimizer selects an Adam optimizer.
Automatic restoration system of calligraphy word stock based on style migration includes: a processor and a storage device; and the processor loads and executes the instructions and data in the storage device to realize the automatic restoration method of the calligraphy font library based on style migration.
Compared with the prior art, the invention has the advantages that:
1. a convolutional neural network is added into the generator, so that relevant information such as strokes and the like can be automatically extracted, and manual image preprocessing is not needed.
2. The Densenet is added into the generator to serve as a conversion module, so that the condition that deformation is large between an input image and an output image can be effectively improved.
3. The Wasserstein distance is added to the loss function. The training process is stable, basically does not need to debug parameters, and is suitable for various calligraphies.
Drawings
the invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of the method for automatically restoring a calligraphy character library based on style migration according to the present invention.
FIG. 2 is a schematic diagram of a generator according to the present invention;
FIG. 3 is a schematic diagram of an arbiter according to the present invention;
FIG. 4 is a diagram illustrating a font repairing effect according to the present invention;
FIG. 5 is a pseudo-code presentation diagram of the Adam optimizer of the present invention.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
The calligraphy font library automatic repairing method based on style migration is shown in figure 1 and comprises a generator, a discriminator and an optimizer, wherein the generator further comprises an encoding module, a decoding module and a converting module; the method comprises the following specific steps:
s1, setting a regular font as an input font, and setting a calligraphy font (hereinafter, referred to as a face font for short) of the Yankanqin, which is an ancient famous calligraphy, of China as a standard style font, wherein the regular font has a complete font library, and the face font has not established the complete font library.
S2, as shown in fig. 2, the regular script chinese character image is used as an input of the generator coding module, and is processed by a series of Convolution-batch norm-leak relu (Convolution-batch normalization-activation function) to obtain the potential feature information (C is depth) of 4 × C of the image. The conversion module converts the latent feature information into a color style feature information using a dense block. And finally, the decoding module processes the converted character information of the color body through a series of Deconvolution-batch norm-LeakyReLU (Deconvolution-batch standardization-activation function) to obtain a Chinese character image of the generated color body, wherein the generated color body possibly has a certain difference with the real color body. The detailed parameters of the network structure of the generator are shown in table 1, and the input font image is 64 × 3 (pixel 64x64, RGB three channels), and becomes 4 × C (pixel 4x4, C channel, which may also be referred to as depth C) after convolution.
TABLE 1 Generator network parameters
Wherein Conv-Norm-Relu represents that the layer network is composed of convolution, batch normalization and activation functions, 4x4 represents that the convolution kernel size stride represents the moving step size, and the filter represents the output channel of the convolution network. The conversion module is composed of a denseblock,
S3, as shown in figure 3, inputting the regular font Chinese character image and the generated color volume Chinese character image into a discriminator (formed by VGG16 network), extracting font characteristic information through 5 layers of convolution of the discriminator, and finally outputting the probability that the generated color volume Chinese character image is a real color volume Chinese character image through three layers of full connection; similarly, the regular script Chinese character image and the standard style color body Chinese character image are used as the input of the discriminator to obtain the probability that the standard style color body Chinese character image is the real standard style color body Chinese character image; finally, obtaining a loss function of the generator and a loss function of the discriminator according to the two probabilities; true means that the style characteristics of the current font image are infinitely close to those of a standard style font.
And S4, combining the feedback results of the loss functions of the generator and the discriminator by the optimizer to adjust the network weights of the generator and the discriminator, and repeatedly executing the steps S2-S4 after adjustment until the loss functions of the generator and the discriminator are converged to obtain the generator after training.
The output of the discriminator is a probability, if the probability is 1, the probability that the Chinese character generated by the generator is true is 100 percent; if the probability is 0, the Chinese character generated by the generator is false. And when the Chinese characters generated by the final generator can be 'falsely and falsely' and cannot be identified by the judgment network, the confrontation training is finished.
inputting a regular script Chinese character image X, wherein the generated color volume Chinese character image is X ', the standard color volume Chinese character image is Y, and D (X') is the probability that the generated picture is true, then:
Loss function of arbiter: l isD＝ln(1-D(X'))+ln(D(Y))，
loss function of generator: l isG＝ln(D(X'))，
The discriminator is to identify the generated picture and the real picture as much as possible, that is, to satisfy the condition that D (X') approaches to 0 and D (Y) approaches to 1, and at this time L is equal to LDApproaching 0.
the generator is to ensure that the generated color volume Chinese character image approaches to the standard color volume Chinese character image, namely to satisfy that D (X') approaches to 1, and at the moment, L isGapproaching 0.
The generator and the discriminator are contradictory and antagonistic, the optimizer adjusts the network weights of the generator and the discriminator to enable the loss functions of the generator and the discriminator to simultaneously tend to zero, and the two networks progress in the antagonistic process.
and S5, inputting all Chinese character images in the regular font library into a generator after training, generating real color font Chinese character images after the generator processing, and obtaining a complete color font library after integrating the color font Chinese character images. As shown in fig. 4, some of the calligraphic words repaired by the present invention can be found to be very close to the real calligraphic words.
The Wasserstein distance is introduced into the loss function, so that the whole training process is more stable, debugging parameters are basically not needed, and the method is suitable for more calligraphy types. Many points in the high-dimensional space are redundant, real data are shrunk on the manifold of the low-dimensional subspace (namely, the high-dimensional curved surface), and because the dimension is low, the occupied space volume is almost 0, so that the original GAN has the problems that the KL distance is always log2 when two distributed support sets are not overlapped or are overlapped very little (the generated data of the generator is widely distributed in the high-dimensional space, the real data cannot be detected), the gradient of the generator disappears (the gradient is always 0, the parameters cannot be optimized, and better pictures cannot be generated), and how to train is useless. The Wasserstein distance (new loss function) replaces the KL divergence of the original GAN (the loss function of the original GAN is calculated from the KL divergence) and is targeted for optimization. Based on the superior smooth characteristic of the Wasserstein distance relative to the KL divergence, the problem of gradient disappearance of the original GAN is fundamentally solved, and the training is more stable. The loss function after introducing the Wasserstein distance is as follows:
Loss function of arbiter:
Loss function of generator:
The wasserstein distance is calculated as Prand PgTwo distributed distances, the image corresponding to a certain distribution, i.e.Is to calculate the difference of the two images. The meaning of the above formula is: for each joint distribution, sampling X ' and Y from the inside and calculating the distance between X ' and Y, then taking all X ' and Y to calculate an expectation (i.e., E in the formula), and taking the distance with the smallest expectation (i.e., inf, the smallest in the set) as the wasserstein distance.
the optimizer selects an Adam optimizer, the Adam optimizer combines the advantages of two optimization algorithms of AdaGrad and RMSProp, First Moment Estimation (namely mean value of gradient) and Second Moment Estimation (namely Second Moment Estimation) of the gradient are comprehensively considered, and the updating step length is calculated. The Adam optimizer has the advantages that: the method is simple to implement, high in calculation efficiency and low in memory requirement; the updating of the parameters is not influenced by the gradient scaling transformation; hyper-parameters are well-interpretable and typically require no or little fine-tuning; the step size of the update can be limited to a rough range (initial learning rate); the step annealing process (automatic adjustment of learning rate) can be naturally realized; the method is very suitable for being applied to large-scale data and parameter scenes; is applicable to unstable objective functions; the method is suitable for the problem of sparse gradient or large noise in the gradient. Pseudo code for the Adam optimizer is shown in fig. 5.
The existing technology can generate calligraphy characters by learning some samples and achieve certain repairing effect. However, there are several problems, mainly embodied in three aspects:
1. The prior art needs complex preprocessing of images, such as stroke extraction, stroke decomposition and labeling. Although much more convenient than manual handling, it is inefficient and still wastes a lot of time and effort.
2. for calligraphy with continuous strokes (such as line writing), stroke information of the calligraphy is difficult to extract, so that the prior art cannot effectively process the calligraphy with the continuous strokes.
3. Generation of the countermeasure network and conditional generation of the countermeasure network are unstable, require careful balancing of the training process of the generator and the arbiter, and lack of diversity in the generated results.
The automatic calligraphy font library repairing method based on style migration is a style migration model based on Pix2Pix and Wasserstein distance. The Pix2Pix is combined with a deep convolutional neural network and a conditional countermeasure generation network, the deep convolutional neural network can directly extract stroke information without additional image preprocessing, gradient disappearance is effectively avoided after the Wasserstein distance, and network training of the method is more stable. Style feature migration is to convert the style of image a into image B, so that B contains both the content of image B and the style of image a. The method does not need to carry out complex preprocessing on font images and adopts a style migration model with stable training process to repair a set of complete calligraphy character library. The method and the device can improve the repairing efficiency, reduce the consumption of manpower and material resources and realize the automation of font repairing.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (4)
1. The calligraphy font library automatic repairing method based on style migration is characterized by comprising a generator, a discriminator, an optimizer, an encoding module, a decoding module and a converting module; the method comprises the following specific steps:
s1, setting a printing font of the existing complete font library as an input font, and setting a calligraphy font to be repaired as a standard style font;
S2, taking the input font image as the input of a coding module in the generator, and carrying out convolution, batch standardization and activation function processing on the input font image by the coding module to obtain the potential characteristic information of the image; the conversion module converts the potential characteristic information into characteristic information of a standard style font by using a dense block; finally, the decoding module carries out deconvolution, batch standardization and activation function processing on the converted characteristic information of the standard style font to obtain a generated font image;
s3, taking the input font image and the generated font image as the input of a discriminator, extracting font characteristic information through 5 layers of convolution of the discriminator, and finally outputting the probability that the generated font image is a real standard style font through three layers of full connection; in the same way, the input font image and the standard style font image are used as the input of the discriminator to obtain the probability that the standard style font image is the real standard style font; finally, obtaining a loss function of the generator and a loss function of the discriminator according to the two probabilities;
S4, the optimizer adjusts the network weights of the generator and the discriminator according to the loss function of the generator and the loss function of the discriminator, and repeatedly executes the steps S2-S4 after adjustment until the loss function of the generator and the loss function of the discriminator are converged to obtain a generator which is trained;
and S5, inputting all Chinese character images in the character library of the input fonts into the generator after training, generating complete Chinese character images with standard style fonts after the generator processing, and obtaining the complete character library with the standard style fonts after the Chinese character images with the standard style fonts are integrated and processed.
2. The method for automatically repairing a calligraphy font library based on style migration according to claim 1, wherein Wasserstein distance is introduced into the loss function.
3. The method for automatically restoring a calligraphy font library based on style migration according to claim 1, wherein the optimizer selects an Adam optimizer.
4. Automatic system of restoreing of calligraphy word stock based on style migration, its characterized in that includes: a processor and a storage device; the processor loads and executes the instructions and data in the storage device to realize the automatic calligraphy word stock repairing method based on style migration as claimed in any one of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910702258.9A CN110570481A (en) | 2019-07-31 | 2019-07-31 | calligraphy word stock automatic repairing method and system based on style migration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910702258.9A CN110570481A (en) | 2019-07-31 | 2019-07-31 | calligraphy word stock automatic repairing method and system based on style migration |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110570481A true CN110570481A (en) | 2019-12-13 |
Family
ID=68773848
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910702258.9A Pending CN110570481A (en) | 2019-07-31 | 2019-07-31 | calligraphy word stock automatic repairing method and system based on style migration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110570481A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956678A (en) * | 2019-12-16 | 2020-04-03 | 北大方正集团有限公司 | Font processing method and device |
| CN111144066A (en) * | 2019-12-27 | 2020-05-12 | 北大方正集团有限公司 | Adjusting method, device and equipment for font of font library and storage medium |
| CN111539873A (en) * | 2020-05-06 | 2020-08-14 | 中国科学院自动化研究所 | Personal style calligraphy character library generation method and system |
| CN112070658A (en) * | 2020-08-25 | 2020-12-11 | 西安理工大学 | Chinese character font style migration method based on deep learning |
| CN112417959A (en) * | 2020-10-19 | 2021-02-26 | 上海臣星软件技术有限公司 | Picture generation method and device, electronic equipment and computer storage medium |
| CN112435196A (en) * | 2020-12-02 | 2021-03-02 | 中国人民解放军战略支援部队信息工程大学 | Character repairing method and system based on deep learning |
| CN112818634A (en) * | 2021-01-29 | 2021-05-18 | 上海海事大学 | Calligraphy work style migration system, method and terminal |
| CN113095038A (en) * | 2021-05-08 | 2021-07-09 | 杭州王道控股有限公司 | Font generation method and device for generating countermeasure network based on multitask discriminator |
| CN113326725A (en) * | 2021-02-18 | 2021-08-31 | 陕西师范大学 | Chinese character font automatic generation method based on framework guide transmission network |
| CN113553932A (en) * | 2021-07-14 | 2021-10-26 | 同济大学 | Calligraphy character erosion repairing method based on style migration |
| CN113724132A (en) * | 2021-11-03 | 2021-11-30 | 浙江宇视科技有限公司 | Image style migration processing method and device, electronic equipment and storage medium |
| CN113761831A (en) * | 2020-11-13 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for generating style calligraphy and storage medium |
| CN114821602A (en) * | 2022-06-28 | 2022-07-29 | 北京汉仪创新科技股份有限公司 | Method, system, apparatus and medium for training an antagonistic neural network to generate a word stock |
| CN115797216A (en) * | 2022-12-14 | 2023-03-14 | 齐鲁工业大学 | Inscription character restoration model and restoration method based on self-coding network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107577651A (en) * | 2017-08-25 | 2018-01-12 | 上海交通大学 | Chinese character style migratory system based on confrontation network |
| CN107644006A (en) * | 2017-09-29 | 2018-01-30 | 北京大学 | A kind of Chinese script character library automatic generation method based on deep neural network |
| CN108304357A (en) * | 2018-01-31 | 2018-07-20 | 北京大学 | A kind of Chinese word library automatic generation method based on font manifold |
| CN109165376A (en) * | 2018-06-28 | 2019-01-08 | 西交利物浦大学 | Style character generating method based on a small amount of sample |
| CN109408776A (en) * | 2018-10-09 | 2019-03-01 | 西华大学 | A kind of calligraphy font automatic generating calculation based on production confrontation network |
| US20190138860A1 (en) * | 2017-11-08 | 2019-05-09 | Adobe Inc. | Font recognition using adversarial neural network training |
-
2019
- 2019-07-31 CN CN201910702258.9A patent/CN110570481A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107577651A (en) * | 2017-08-25 | 2018-01-12 | 上海交通大学 | Chinese character style migratory system based on confrontation network |
| CN107644006A (en) * | 2017-09-29 | 2018-01-30 | 北京大学 | A kind of Chinese script character library automatic generation method based on deep neural network |
| US20190138860A1 (en) * | 2017-11-08 | 2019-05-09 | Adobe Inc. | Font recognition using adversarial neural network training |
| CN108304357A (en) * | 2018-01-31 | 2018-07-20 | 北京大学 | A kind of Chinese word library automatic generation method based on font manifold |
| CN109165376A (en) * | 2018-06-28 | 2019-01-08 | 西交利物浦大学 | Style character generating method based on a small amount of sample |
| CN109408776A (en) * | 2018-10-09 | 2019-03-01 | 西华大学 | A kind of calligraphy font automatic generating calculation based on production confrontation network |
Non-Patent Citations (2)
| Title |
|---|
| MENGXI QIN等: "Restore the Incomplete Calligraphy Based on Style Transfer", 《PROCEEDINGS OF THE 38TH CHINESE CONTROL CONFERENCE》 * |
| 白海娟等: "基于生成式对抗网络的字体风格迁移方法", 《大连民族大学学报》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956678A (en) * | 2019-12-16 | 2020-04-03 | 北大方正集团有限公司 | Font processing method and device |
| CN111144066B (en) * | 2019-12-27 | 2022-02-18 | 北大方正集团有限公司 | Adjusting method, device and equipment for font of font library and storage medium |
| CN111144066A (en) * | 2019-12-27 | 2020-05-12 | 北大方正集团有限公司 | Adjusting method, device and equipment for font of font library and storage medium |
| CN111539873A (en) * | 2020-05-06 | 2020-08-14 | 中国科学院自动化研究所 | Personal style calligraphy character library generation method and system |
| CN111539873B (en) * | 2020-05-06 | 2023-10-20 | 中国科学院自动化研究所 | Method and system for generating personal style handwriting word stock |
| CN112070658A (en) * | 2020-08-25 | 2020-12-11 | 西安理工大学 | Chinese character font style migration method based on deep learning |
| CN112070658B (en) * | 2020-08-25 | 2024-04-16 | 西安理工大学 | Deep learning-based Chinese character font style migration method |
| CN112417959A (en) * | 2020-10-19 | 2021-02-26 | 上海臣星软件技术有限公司 | Picture generation method and device, electronic equipment and computer storage medium |
| CN113761831A (en) * | 2020-11-13 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for generating style calligraphy and storage medium |
| CN112435196B (en) * | 2020-12-02 | 2022-11-29 | 中国人民解放军战略支援部队信息工程大学 | Character repairing method and system based on deep learning |
| CN112435196A (en) * | 2020-12-02 | 2021-03-02 | 中国人民解放军战略支援部队信息工程大学 | Character repairing method and system based on deep learning |
| CN112818634B (en) * | 2021-01-29 | 2024-04-05 | 上海海事大学 | Handwriting style migration system, method and terminal |
| CN112818634A (en) * | 2021-01-29 | 2021-05-18 | 上海海事大学 | Calligraphy work style migration system, method and terminal |
| CN113326725A (en) * | 2021-02-18 | 2021-08-31 | 陕西师范大学 | Chinese character font automatic generation method based on framework guide transmission network |
| CN113326725B (en) * | 2021-02-18 | 2024-03-12 | 陕西师范大学 | Automatic Chinese character font generating method based on skeleton guiding transmission network |
| CN113095038B (en) * | 2021-05-08 | 2024-04-16 | 杭州王道控股有限公司 | Font generation method and device for generating countermeasure network based on multi-task discriminator |
| CN113095038A (en) * | 2021-05-08 | 2021-07-09 | 杭州王道控股有限公司 | Font generation method and device for generating countermeasure network based on multitask discriminator |
| CN113553932B (en) * | 2021-07-14 | 2022-05-13 | 同济大学 | Calligraphy character erosion repairing method based on style migration |
| CN113553932A (en) * | 2021-07-14 | 2021-10-26 | 同济大学 | Calligraphy character erosion repairing method based on style migration |
| CN113724132A (en) * | 2021-11-03 | 2021-11-30 | 浙江宇视科技有限公司 | Image style migration processing method and device, electronic equipment and storage medium |
| CN113724132B (en) * | 2021-11-03 | 2022-02-18 | 浙江宇视科技有限公司 | Image style migration processing method and device, electronic equipment and storage medium |
| CN114821602A (en) * | 2022-06-28 | 2022-07-29 | 北京汉仪创新科技股份有限公司 | Method, system, apparatus and medium for training an antagonistic neural network to generate a word stock |
| CN115797216A (en) * | 2022-12-14 | 2023-03-14 | 齐鲁工业大学 | Inscription character restoration model and restoration method based on self-coding network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110570481A (en) | calligraphy word stock automatic repairing method and system based on style migration | |
| CN106384094B (en) | A kind of Chinese word library automatic generation method based on writing style modeling | |
| CN107239801B (en) | Video attribute representation learning method and video character description automatic generation method | |
| CN108629338B (en) | Face beauty prediction method based on LBP and convolutional neural network | |
| CN108764242A (en) | Off-line Chinese Character discrimination body recognition methods based on deep layer convolutional neural networks | |
| Tian et al. | Kaokore: A pre-modern japanese art facial expression dataset | |
| CN112163401A (en) | Chinese character font generation method based on compressed and excited GAN network | |
| CN111666937A (en) | Method and system for recognizing text in image | |
| CN111062329A (en) | Unsupervised pedestrian re-identification method based on augmented network | |
| CN108537109B (en) | OpenPose-based monocular camera sign language identification method | |
| CN112381082A (en) | Table structure reconstruction method based on deep learning | |
| CN113095421A (en) | Method for generating font database, and training method and device of neural network model | |
| CN110851627B (en) | Method for describing sun black subgroup in full-sun image | |
| CN115170403A (en) | Font repairing method and system based on deep meta learning and generation countermeasure network | |
| CN113421318B (en) | Font style migration method and system based on multitask generation countermeasure network | |
| Vinokurov | Using a convolutional neural network to recognize text elements in poor quality scanned images | |
| CN105069767A (en) | Image super-resolution reconstruction method based on representational learning and neighbor constraint embedding | |
| CN109493279B (en) | Large-scale unmanned aerial vehicle image parallel splicing method | |
| Prabhu et al. | Fonts-2-handwriting: A seed-augment-train framework for universal digit classification | |
| CN111242216A (en) | Image generation method for generating anti-convolution neural network based on conditions | |
| Sasipriyaa et al. | Recognizing handwritten offline Tamil character by using cGAN & CNN | |
| CN113052759B (en) | Scene complex text image editing method based on MASK and automatic encoder | |
| CN112132916B (en) | Seal cutting work customized design generating device for generating countermeasure network | |
| CN112163605A (en) | Multi-domain image translation method based on attention network generation | |
| Zhang et al. | Calligraphy fonts generation based on generative adversarial networks |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191213 |
|
| RJ01 | Rejection of invention patent application after publication |