CN108764070A - A kind of stroke dividing method and calligraphic copying guidance method based on writing video - Google Patents
A kind of stroke dividing method and calligraphic copying guidance method based on writing video Download PDFInfo
- Publication number
- CN108764070A CN108764070A CN201810446094.3A CN201810446094A CN108764070A CN 108764070 A CN108764070 A CN 108764070A CN 201810446094 A CN201810446094 A CN 201810446094A CN 108764070 A CN108764070 A CN 108764070A
- Authority
- CN
- China
- Prior art keywords
- video
- stroke
- writing
- convolution
- group
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes
-
- 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
Abstract
The invention discloses a kind of based on the stroke dividing method and calligraphic copying guidance method of writing video, wherein, based on the stroke dividing method for writing video, redesign convolutional neural networks, network and Recognition with Recurrent Neural Network are combined by convolution god, the parameter for greatly reducing neural network, accelerates training speed；And the ability to express to the space characteristics of video frame is enhanced, while being extracted the temporal motion information of writing brush word writing again, realizes the fine granularity action recognition of high-accuracy.The calligraphic copying guidance method of the present invention, using above-mentioned based on the stroke dividing method for writing video, the stroke video divided using stroke can realize the accurate guidance to calligraphic copying process.
Description
Technical field
The invention belongs to Computer Recognition Technology fields, are related to a kind of based on the stroke dividing method and calligraphy of writing video
Copy guidance method.
Background technology
Chinese calligraphy is a kind of unique visual art, is thousands of years of Sinitic crystallizations, to traditional Chinese culture
Succession plays an important role.Gradual attention with China to its traditional culture, more and more Chinese and the foreigner
Start to learn Chinese calligraphy.And the basis of learning calligraphy be grasp writing process in the method wield the pen, it is just so-called " calligraphy it
It is wonderful, wieling the pen entirely " (Yu Chu Qing Dynastys Kang Youwei), and also there is calligraphist Cai river in Guangxi to write in ancient times《The style of brushwork》One book reflects that calligraphy is transported
Importance of the pen in Chinese calligraphy's writing process.
The common learning method of calligraphy beginner is to copy study, in the process, it is necessary first to which grasp is exactly basic
The pen manipulating method of stroke.However, at present Chinese calligraphy there is a serious shortage of teaching resource, cause many calligraphy learners be difficult to obtain it is excellent
The calligraphy teaching of matter, this will be as serious one of the principal element for hindering the study of Chinese calligraphy to promote.In order to learning calligraphy
Person instructs, and is realized using the method that calligraphy action is identified.
Lack in the prior art and (still image or dynamic are either directed to the related work that calligraphy action of writing identifies
Video).Similar work has based on image or the human body behavior of video or the identification of action therewith, includes mainly conventional machines
Learning method and deep learning method.With the extensive use of deep learning method, it is largely based on convolutional neural networks (CNN)
Method is used for the action recognition in static video frame, and relative to conventional machines learning method, recognition effect is significantly improved.
Time convolution is added in Andrej et al. in the frame of CNN, to obtain the local motion information in time domain, relative to static CNN
Recognition methods improve 2%.But since the time convolution can only obtain the time-domain information of of short duration several frames, and it is small in motion amplitude
Video in, the static information of a few frame images is almost consistent, it is difficult to achieve the purpose that obtain temporal motion information, therefore such side
Method changes action subtle video, and accuracy of identification is limited.
Invention content
For problems of the prior art, the object of the present invention is to provide a kind of based on the stroke for writing video
Dividing method, this method can accurately divide the work of starting writing and start to write of writer, and realize the purpose of stroke segmentation.
To achieve the goals above, the present invention uses following technical scheme：
It is a kind of based on write video stroke dividing method, this method for will write video be divided into according to stroke it is multiple
Stroke video, includes the following steps：
Step 1, the writing video for writing single word is obtained, which includes multiple image；By the multiple image point
At multiple video groups, each video group includes the continuous image of n frames；
Step 2, for each video group, all images that video group includes are input in convolutional neural networks, convolution
Neural network exports the corresponding image space feature vector of the video group；The convolutional neural networks include two the first convolution groups,
Two the second convolution groups, three full articulamentums and one softmax layers, wherein the first convolution group includes according to transmission side data
To sequentially connected two convolutional layers and a pond layer, the second convolution group includes according to data transfer direction sequentially connected three
A convolutional layer and a pond layer；Three full articulamentums and one softmax layers are sequentially connected according to data transfer direction, three
Last four layer for convolutional neural networks of full articulamentum and one softmax layer；
Step 3, the corresponding image space feature vector of video group is input in Recognition with Recurrent Neural Network, Recognition with Recurrent Neural Network
The state for exporting the video group is writing state or non-writing state；
Step 4, two adjacent video groups, the phase of the selection are chosen in all video groups that state is non-writing state
All video groups between two adjacent video groups are grouped together into a stroke video；State for non-writing state institute
All video groups before having first in video group video group are grouped together into a stroke video；State is non-book
All video groups after the last one video group in all video groups of write state are grouped together into a stroke and regard
Frequently.
Optionally, described two first convolution groups, two the second convolution groups, three full articulamentums and one softmax layers
It is sequentially connected according to data transfer direction.
Optionally, the convolution window of described two first convolution groups and all convolutional layers in two the second convolution groups is big
Small is 3 × 3, and the pond window of all pond layers is 2 × 2；
The number of the convolution kernel of convolutional layer in first the first convolution group is 64, the convolutional layer of second the first convolution group
The number of convolution kernel be 128, the number of the convolution kernel of the convolutional layer in first the second convolution group is 256, second second
The number of the convolution kernel of the convolutional layer of convolution group is 512.
The present invention also provides a kind of calligraphic copying guidance methods, include the following steps：
Step 1, by the writing video of the writing video of copier and standard be divided into respectively multiple imitation stroke videos and
Multiple standard stroke videos；Wherein, the writing video of copier and the writing video that the writing video of standard is the same word are more
A imitation stroke video and multiple standard stroke videos are corresponded according to stroke；
Step 2, it is all made of trace tracking method to copying stroke video and standard stroke video and handles, respectively obtain
Copy the corresponding trajectory coordinates sequence of stroke video and the corresponding trajectory coordinates sequence of standard stroke video；
Step 3, using dynamic time warping method calculate copy stroke video trajectory coordinates sequence and with the copying pen
Draw the similarity between the trajectory coordinates sequence of the corresponding standard stroke video of video；
Step 4, all similarities that step 3 obtains combine to form a feature vector, and this feature vector is input to line
Property regression model in, the linear regression model (LRM) export score value；
The writing video by the writing video of copier and standard in the step 1 is divided into multiple imitation strokes respectively
Video and multiple standard stroke videos are obtained according to above-mentioned based on the stroke dividing method for writing video.
Compared with prior art, the present invention has the following technical effects：The present invention's is divided based on the stroke for writing video
Method redesigns convolutional neural networks, network and Recognition with Recurrent Neural Network is combined by convolution god, greatly reduce nerve net
The parameter of network, accelerates training speed；And the ability to express to the space characteristics of video frame is enhanced, while being extracted book again
The temporal motion information that method word is write, realizes the fine granularity action recognition of high-accuracy.The calligraphic copying guidance method of the present invention,
Using above-mentioned based on the stroke dividing method for writing video, the stroke video divided using stroke can be realized to calligraphy
The accurate guidance of imitation process.
Explanation and illustration in further detail is made to the solution of the present invention with reference to the accompanying drawings and detailed description.
Description of the drawings
Fig. 1 is the flow chart based on the stroke dividing method for writing video of the present invention.
Specific implementation mode
The invention discloses a kind of based on the stroke dividing method for writing video, and referring to Fig. 1, this method is regarded for that will write
Frequency is divided into multiple stroke videos according to stroke, specifically includes following steps
Step 1, the writing video for writing single word is obtained, which includes multiple image；By the multiple image point
At multiple video groups, each video group includes the continuous image of n frames, in the present embodiment, n=5.
Step 2, for each video group, all images that video group includes are input in convolutional neural networks, convolution
Neural network exports the corresponding image space feature vector of the video group；The convolutional neural networks include two the first convolution groups,
Two the second convolution groups, three full articulamentums and one softmax layers, wherein the first convolution group includes according to transmission side data
To sequentially connected two convolutional layers and a pond layer, the second convolution group includes according to data transfer direction sequentially connected three
A convolutional layer and a pond layer；Three full articulamentums and one softmax layers are sequentially connected according to data transfer direction, three
Last four layer for convolutional neural networks of full articulamentum and one softmax layer.
Step 3, the corresponding image space feature vector of video group is input in Recognition with Recurrent Neural Network, Recognition with Recurrent Neural Network
The state for exporting the video group is writing state or non-writing state.
Step 4, two adjacent video groups, the phase of the selection are chosen in all video groups that state is non-writing state
All video groups between two adjacent video groups are grouped together into a stroke video；State for non-writing state institute
All video groups before having first in video group video group are grouped together into a stroke video；State is non-book
All video groups after the last one video group in all video groups of write state are grouped together into a stroke and regard
Frequently.
Big convolution kernel is decomposed into several small-sized convolution kernels and is superimposed by the present invention.Use the volume of multiple smaller convolution kernels
The convolutional layer that lamination replaces a convolution kernel larger so that the level of network is deeper, can not only reduce parameter, and carry out
More Nonlinear Mappings, can increase the fitting ability to express of network, can not only extract the mass motion information wield the pen, and
And the minutia of the vigour of style in writing can also be extracted.
Preferably, in another embodiment, two the first convolution groups, two the second convolution groups, three full articulamentums and one
A softmax layers is sequentially connected according to data transfer direction.Traditional big window is replaced by the convolutional layer combination of multiple wickets
Mouth convolution so that the level of network is deeper, stronger to the space characteristics ability to express of subtle posture of wieling the pen in calligraphy, another party
Face is effectively reduced the parameter of whole network, more efficient.
Optionally, in another embodiment, two the first convolution groups and all convolutional layers in two the second convolution groups
The size of convolution window is 3 × 3, and the pond window of all pond layers is 2 × 2；Volume in first the first convolution group
The number of the convolution kernel of lamination is 64, and the number of the convolution kernel of the convolutional layer of second the first convolution group is 128, first second
The number of the convolution kernel of convolutional layer in convolution group is 256, and the number of the convolution kernel of the convolutional layer of second the second convolution group is
512。
The invention also discloses a kind of calligraphic copying guidance methods, include the following steps：
Step 1, by the writing video of the writing video of copier and standard be divided into respectively multiple imitation stroke videos and
Multiple standard stroke videos；Wherein, the writing video of copier and the writing video that the writing video of standard is the same word are more
A imitation stroke video and multiple standard stroke videos are corresponded according to stroke.
Step 2, it is all made of trace tracking method to copying stroke video and standard stroke video and handles, respectively obtain
Copy the corresponding trajectory coordinates sequence of stroke video and the corresponding trajectory coordinates sequence of standard stroke video；Track following herein
Method uses TLD track following algorithms.
Step 3, it is calculated using dynamic time warping method (DTW) and copies the trajectory coordinates sequence of stroke video and face with this
Imitate the similarity between the trajectory coordinates sequence of the corresponding standard stroke video of stroke video.
Step 4, all similarities that step 3 obtains combine to form a feature vector, and this feature vector is input to line
Property regression model in, the linear regression model (LRM) export score value.
It is divided into multiple imitation strokes to regard respectively the writing video of the writing video of copier and standard in above-mentioned steps 1
Frequency and multiple standard stroke videos are obtained according to above-mentioned based on the stroke dividing method for writing video.
Embodiment
The present embodiment establishes calligraphy and writes video database, and based on the database to the method for the present invention, with
And existing method is tested.The data source of calligraphy video frequency database is mainly the Devil's training of Zhejiang University teacher Yu Zhonghua
1 writing master and 6 students of battalion are in the HD video recorded when calligraphic copying writing.It recorded this 7 calligraphies altogether
Writer writes the Yan style regular script of Chinese characters such as " ten, ancient, greatly, husband are interior, on, or not brother is left, son " respectively, this ten words are students
Do constant practice the Chinese character of writing when starting to practice calligraphy, and per each word of person writing 20 times, therefore the segment that complete word is write
Total to have 7*10*20=1400 complete individual characters to write video-frequency band, the resolution ratio of video frame is 1920*1080, and frame rate is
50 frames/second.Based on these video datas, calligraphy stroke is carried out respectively and is divided into stroke writing sub-video, TLD carries out writing brush track
Tracking, DTW carries out the conversion of writing brush track characteristic, and linear regression carries out the tasks such as assessment of calligraphy track.
(1) Video segmentation will be write into stroke video
More than 1400 a videos of 10 Chinese characters are applied with the method (MCNN-LSTM) of the present invention, CNN and CNN-LSTM respectively
The switching time point between each stroke to identify each complete word, in the writing process in these Hanzhong, one start to write with
One start writing between process be exactly to write the process of a stroke；Then according to the result of stroke segmentation by the video of complete word
It is divided into the sub-video of several strokes.Wherein training data is 300 videos of random selection from 1400 videos, is then marked
Remember start to write sequence of frames of video, writing process frame sequence and frame sequence of starting writing, other remaining videos are as test data.It is testing
In, the input of CNN is single frame data, carries out video frame identification, the input of CNN-LSTM and MCNN-LSTM are continuous 5 frames
Frame sequence, carries out the identification of continuous frame sequence, and the action action frame recognition sequence started writing or started to write.The convolutional layer knot of MCNN-LSTM
Structure designs as it was noted above, LSTM hidden layers use 256 neurons, equally using stochastic gradient descent method using stochastic gradient
Descent method, bath size are 50, learning rate 0.001.The results are shown in Table 1 for test experiments：
It can be seen from the data in Table 1 that the discrimination of CNN has two row data, it is single that reason is that this CNN is mainly carried out
Frame identification, therefore the first row is using about 10,000 frames as training, frame discrimination of remaining 30,000 frame as test.And the
Two, three, four rows are checked by frame tagging, the discrimination for the continuous frame sequence being calculated, and the second row is five in continuous five frame
Frame entirely to video-frequency band discrimination, as long as the third line is that have 4 frames to identify correct video-frequency band discrimination, fourth line in continuous five frame
As long as being to there are 3 frames to identify correct video-frequency band discrimination in continuous five frame.From experimental data it can be seen that simple convolutional Neural
Network is difficult to obtain preferable effect in the video-frequency band identification with timing information, and 60% frame identifies just only in successive frame
True video-frequency band discrimination is just close to the recognition result of the LSTM with temporal aspect.In contrast, continuous frame sequence is passed through
Input of the feature that convolutional neural networks obtain as LSTM, the identification for then carrying out video-frequency band can on the basis of original
Obtain prodigious performance boost, it is seen that LSTM is in the signal processing method with temporal aspect with excellent characteristic.And it compares
Under, the MCNN designed herein has the small convolution kernel superposition of multilayer, relatively simple big convolution kernel or simple small convolution kernel,
MCNN can preferably capture details and Global movement feature, then it is combined with LSTM.Therefore, MCNN-LSTM can be obtained
The recognition effect of better sequence of frames of video.Although from experimental data as can be seen that MCNN-LSTM in 10 words with its other party
The discrimination difference of method is irregular, but is generally completely better than control methods.
Table 1
(2) calligraphic copying is instructed
After being divided into stroke sub-video by stroke to the video of complete word using MCNN-LSTM, Faster is recycled
RCNN-TLD carries out the tracking of writing brush track in these sub-videos, obtains everyone in the track for each stroke for writing each word
Information.Then the trace information is converted into the required interdependent similarity characteristic information in track in linear regression model (LRM), and tied
Calligraphy teacher is closed to the scoring of each student's written word process to obtain everyone score when writing each stroke and write whole
The score of a word.Wherein table 2 is shown six students scoring of each stroke and scoring of entire word at writing " big " and (is somebody's turn to do
Scoring is that everyone writes the mean value of 20 times scorings) with the comparison manually scored, the evaluation of other words is similar therewith.This is manually commented
It is their teacher to divide, i.e. evaluation of the template writer to every student's writing process.As can be seen from the table, it carries herein
The evaluation system gone out substantially conforms to manually evaluate.It lays the foundation to the automatic Evaluation of calligraphy writing process for the later stage.
Table 2
Claims (4)
1. a kind of based on the stroke dividing method for writing video, this method is divided into multiple pens for that will write video according to stroke
Draw video, which is characterized in that include the following steps：
Step 1, the writing video for writing single word is obtained, which includes multiple image；The multiple image is divided into more
A video group, each video group include the continuous image of n frames；
Step 2, for each video group, all images that video group includes are input in convolutional neural networks, convolutional Neural
Network exports the corresponding image space feature vector of the video group；The convolutional neural networks include two the first convolution groups, two
Second convolution group, three full articulamentums and one softmax layers, wherein the first convolution group include according to data transfer direction according to
Two convolutional layers and a pond layer of secondary connection, the second convolution group include according to sequentially connected three volumes of data transfer direction
Lamination and a pond layer；Three full articulamentums and one softmax layers are sequentially connected according to data transfer direction, and three connect entirely
Connect last four layers that layer and one softmax layers are convolutional neural networks；
Step 3, the corresponding image space feature vector of video group is input in Recognition with Recurrent Neural Network, Recognition with Recurrent Neural Network output
The state of the video group is writing state or non-writing state；
Step 4, two adjacent video groups are chosen in all video groups that state is non-writing state, the selection it is adjacent
All video groups between two video groups are grouped together into a stroke video；State is that all of non-writing state regard
All video groups before first video group in frequency group are grouped together into a stroke video；State is non-writing shape
All video groups after the last one video group in all video groups of state are grouped together into a stroke video.
2. as described in claim 1 based on the stroke dividing method for writing video, which is characterized in that described two first convolution
Group, two the second convolution groups, three full articulamentums and one softmax layers are sequentially connected according to data transfer direction.
3. as claimed in claim 2 based on the stroke dividing method for writing video, which is characterized in that described two first convolution
The size of the convolution window of group and all convolutional layers in two the second convolution groups is 3 × 3, the pond window of all pond layers
Mouth is 2 × 2；
The number of the convolution kernel of convolutional layer in first the first convolution group is 64, the volume of the convolutional layer of second the first convolution group
The number of product core is 128, and the number of the convolution kernel of the convolutional layer in first the second convolution group is 256, second the second convolution
The number of the convolution kernel of the convolutional layer of group is 512.
4. a kind of calligraphic copying guidance method, includes the following steps：
Step 1, the writing video of the writing video of copier and standard is divided into multiple imitation stroke videos and multiple respectively
Standard stroke video；Wherein, the writing video for writing video and standard of copier is the writing video of the same word, Duo Gelin
It imitates stroke video and multiple standard stroke videos is corresponded according to stroke；
Step 2, it is all made of trace tracking method to copying stroke video and standard stroke video and handles, respectively obtain imitation
The corresponding trajectory coordinates sequence of stroke video and the corresponding trajectory coordinates sequence of standard stroke video；
Step 3, it is calculated using dynamic time warping method and copies the trajectory coordinates sequence of stroke video and regarded with the imitation stroke
Frequently the similarity between the trajectory coordinates sequence of corresponding standard stroke video；
Step 4, all similarities that step 3 obtains combine to form a feature vector, and this feature vector is input to linear return
Return in model, which exports score value；
The writing video by the writing video of copier and standard in the step 1 is divided into multiple imitation stroke videos respectively
With multiple standard stroke videos, being divided based on the stroke for writing video according to any claim in claim 1-3
Method obtains.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810446094.3A CN108764070B (en) | 2018-05-11 | 2018-05-11 | Stroke segmentation method based on writing video and calligraphy copying guidance method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810446094.3A CN108764070B (en) | 2018-05-11 | 2018-05-11 | Stroke segmentation method based on writing video and calligraphy copying guidance method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108764070A true CN108764070A (en) | 2018-11-06 |
| CN108764070B CN108764070B (en) | 2021-12-31 |
Family
ID=64009480
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810446094.3A Expired - Fee Related CN108764070B (en) | 2018-05-11 | 2018-05-11 | Stroke segmentation method based on writing video and calligraphy copying guidance method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108764070B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685115A (en) * | 2018-11-30 | 2019-04-26 | 西北大学 | A kind of the fine granularity conceptual model and learning method of bilinearity Fusion Features |
| CN109918991A (en) * | 2019-01-09 | 2019-06-21 | 天津科技大学 | Soft pen calligraphy based on deep learning copies evaluation method |
| CN110503101A (en) * | 2019-08-23 | 2019-11-26 | 北大方正集团有限公司 | Font evaluation method, device, equipment and computer readable storage medium |
| CN111081117A (en) * | 2019-05-10 | 2020-04-28 | 广东小天才科技有限公司 | Writing detection method and electronic equipment |
| CN111477040A (en) * | 2020-05-19 | 2020-07-31 | 西北大学 | Induced calligraphy training system, equipment and method |
| CN111738330A (en) * | 2020-06-19 | 2020-10-02 | 电子科技大学中山学院 | Intelligent automatic scoring method for hand-drawn copy works |
| CN112001236A (en) * | 2020-07-13 | 2020-11-27 | 上海翎腾智能科技有限公司 | Writing behavior identification method and device based on artificial intelligence |
| CN114530064A (en) * | 2022-02-22 | 2022-05-24 | 北京思明启创科技有限公司 | Calligraphy practicing method, calligraphy practicing device, calligraphy practicing equipment and calligraphy practicing storage medium based on video |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4653107A (en) * | 1983-12-26 | 1987-03-24 | Hitachi, Ltd. | On-line recognition method and apparatus for a handwritten pattern |
| CN1268691A (en) * | 2000-04-06 | 2000-10-04 | 许雄 | Intelligent method for handwriting copying |
| CN1658221A (en) * | 2004-01-14 | 2005-08-24 | 国际商业机器公司 | Method and apparatus for performing handwriting recognition by analysis of stroke start and end points |
| CN103226388A (en) * | 2013-04-07 | 2013-07-31 | 华南理工大学 | Kinect-based handwriting method |
| CN104793724A (en) * | 2014-01-16 | 2015-07-22 | 北京三星通信技术研究有限公司 | Sky-writing processing method and device |
| CN104834890A (en) * | 2015-02-13 | 2015-08-12 | 浙江大学 | Method for extracting expression information of characters in calligraphy work |
| CN106095104A (en) * | 2016-06-20 | 2016-11-09 | 电子科技大学 | Continuous gesture path dividing method based on target model information and system |
| CN107067031A (en) * | 2017-03-29 | 2017-08-18 | 西北大学 | A kind of calligraphy posture automatic identifying method based on Wi Fi signals |
| CN107195220A (en) * | 2017-05-27 | 2017-09-22 | 广东小天才科技有限公司 | Writing and learning method, writing learning device and electric terminal |
| CN107704788A (en) * | 2017-09-22 | 2018-02-16 | 西北大学 | A kind of calligraphic copying method based on RF technologies |
-
2018
- 2018-05-11 CN CN201810446094.3A patent/CN108764070B/en not_active Expired - Fee Related
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4653107A (en) * | 1983-12-26 | 1987-03-24 | Hitachi, Ltd. | On-line recognition method and apparatus for a handwritten pattern |
| CN1268691A (en) * | 2000-04-06 | 2000-10-04 | 许雄 | Intelligent method for handwriting copying |
| CN1658221A (en) * | 2004-01-14 | 2005-08-24 | 国际商业机器公司 | Method and apparatus for performing handwriting recognition by analysis of stroke start and end points |
| CN103226388A (en) * | 2013-04-07 | 2013-07-31 | 华南理工大学 | Kinect-based handwriting method |
| CN104793724A (en) * | 2014-01-16 | 2015-07-22 | 北京三星通信技术研究有限公司 | Sky-writing processing method and device |
| CN104834890A (en) * | 2015-02-13 | 2015-08-12 | 浙江大学 | Method for extracting expression information of characters in calligraphy work |
| CN106095104A (en) * | 2016-06-20 | 2016-11-09 | 电子科技大学 | Continuous gesture path dividing method based on target model information and system |
| CN107067031A (en) * | 2017-03-29 | 2017-08-18 | 西北大学 | A kind of calligraphy posture automatic identifying method based on Wi Fi signals |
| CN107195220A (en) * | 2017-05-27 | 2017-09-22 | 广东小天才科技有限公司 | Writing and learning method, writing learning device and electric terminal |
| CN107704788A (en) * | 2017-09-22 | 2018-02-16 | 西北大学 | A kind of calligraphic copying method based on RF technologies |
Non-Patent Citations (6)
| Title |
|---|
| JEFF DONAHUE 等: "Long-term Recurrent Convolutional Networks for Visual Recognition and Description", 《ARXIV:1411.4389V4》 * |
| XIAOOU TANG 等: "Video-Based Handwritten Chinese Character Recognition", 《IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY》 * |
| YUANDONG SUN 等: "A geometric approach to stroke extraction for the Chinese calligraphy robot", 《2014 IEEE INTERNATIONAL CONFERENCE ON ROBOTICS AND AUTOMATION (ICRA)》 * |
| YUANDONG SUN 等: "Robot learns Chinese calligraphy from Demonstrations", 《ENGINEERING, COMPUTER SCIENCE 2014 IEEE/RSJ INTERNATIONAL CONFERENCE ON INTELLIGENT ROBOTS AND SYSTEMS》 * |
| 夏洋: "基于本体模型的特定风格书法字合成研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 熊鹏: "汉字笔迹的笔划提取", 《万方数据知识服务平台》 * |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109685115A (en) * | 2018-11-30 | 2019-04-26 | 西北大学 | A kind of the fine granularity conceptual model and learning method of bilinearity Fusion Features |
| CN109918991A (en) * | 2019-01-09 | 2019-06-21 | 天津科技大学 | Soft pen calligraphy based on deep learning copies evaluation method |
| CN111081117A (en) * | 2019-05-10 | 2020-04-28 | 广东小天才科技有限公司 | Writing detection method and electronic equipment |
| CN110503101A (en) * | 2019-08-23 | 2019-11-26 | 北大方正集团有限公司 | Font evaluation method, device, equipment and computer readable storage medium |
| CN111477040A (en) * | 2020-05-19 | 2020-07-31 | 西北大学 | Induced calligraphy training system, equipment and method |
| CN111477040B (en) * | 2020-05-19 | 2021-07-20 | 西北大学 | Induced calligraphy training system, equipment and method |
| CN111738330A (en) * | 2020-06-19 | 2020-10-02 | 电子科技大学中山学院 | Intelligent automatic scoring method for hand-drawn copy works |
| CN112001236A (en) * | 2020-07-13 | 2020-11-27 | 上海翎腾智能科技有限公司 | Writing behavior identification method and device based on artificial intelligence |
| CN114530064A (en) * | 2022-02-22 | 2022-05-24 | 北京思明启创科技有限公司 | Calligraphy practicing method, calligraphy practicing device, calligraphy practicing equipment and calligraphy practicing storage medium based on video |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108764070B (en) | 2021-12-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108764070A (en) | A kind of stroke dividing method and calligraphic copying guidance method based on writing video | |
| Yang et al. | Chinese character-level writer identification using path signature feature, DropStroke and deep CNN | |
| Lian et al. | EasyFont: a style learning-based system to easily build your large-scale handwriting fonts | |
| CN107766842A (en) | A kind of gesture identification method and its application | |
| CN106845525A (en) | A kind of depth confidence network image bracket protocol based on bottom fusion feature | |
| CN104166499A (en) | Handwriting practice system and practice handwriting automatic detecting and evaluating method | |
| CN104050453A (en) | Evaluation method for handwritten Chinese character handwriting | |
| CN107067031A (en) | A kind of calligraphy posture automatic identifying method based on Wi Fi signals | |
| Li et al. | Neural abstract style transfer for chinese traditional painting | |
| Xu et al. | Evaluating Brush Movements for Chinese Calligraphy: A Computer Vision Based Approach. | |
| Qiao et al. | Efficient style-corpus constrained learning for photorealistic style transfer | |
| CN103544468B (en) | 3D facial expression recognizing method and device | |
| Johnson et al. | Detecting pianist hand posture mistakes for virtual piano tutoring | |
| CN103336830B (en) | Image search method based on structure semantic histogram | |
| CN114676256A (en) | Text classification method based on multi-teaching-assistant model knowledge distillation training | |
| Yuan et al. | Learning to compose stylistic calligraphy artwork with emotions | |
| Sun | Design and Construction of University Book Layout Based on Text Image Preprocessing Algorithm in Education Metaverse Environment | |
| Ma et al. | A deep learning approach for online learning emotion recognition | |
| Yang et al. | Handwriting posture prediction based on unsupervised model | |
| Raut et al. | Generative Adversarial Networks and Deep Learning: Theory and Applications | |
| CN111783697A (en) | Wrong question detection and target recommendation system and method based on convolutional neural network | |
| Husain et al. | The relevance of Freeman Chain Code for copying activities | |
| Chen et al. | Data‐driven Handwriting Synthesis in a Conjoined Manner | |
| Sukamto et al. | Learners mood detection using Convolutional Neural Network (CNN) | |
| Qiang et al. | Research on Hard-tipped Calligraphy Classification Based on Deep Learning Method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20211231 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |