CN104992143B - A kind of Chinese-character stroke extraction method towards vector font - Google Patents
A kind of Chinese-character stroke extraction method towards vector font Download PDFInfo
- Publication number
- CN104992143B CN104992143B CN201510303067.7A CN201510303067A CN104992143B CN 104992143 B CN104992143 B CN 104992143B CN 201510303067 A CN201510303067 A CN 201510303067A CN 104992143 B CN104992143 B CN 104992143B
- Authority
- CN
- China
- Prior art keywords
- stroke
- point set
- point
- ownership
- font
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/22—Character recognition characterised by the type of writing
- G06V30/226—Character recognition characterised by the type of writing of cursive writing
- G06V30/2268—Character recognition characterised by the type of writing of cursive writing using stroke segmentation
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Controls And Circuits For Display Device (AREA)
Abstract
The present invention relates to a kind of Chinese-character stroke extraction method towards vector font, step includes：1) skeletal extraction is carried out to the template font with stroke categorizing information in target font to be extracted and template database corresponding thereto, obtains data point set and template point set；2) using the stroke attaching relation of data point set, template point set and template point set as input, the non-rigid point set registration based on structural information is carried out, the stroke attaching relation of data point set is obtained；3) the stroke attaching relation of data point set is converted into the attaching relation of data profile section, makes each contour segment of composition character contour that there is corresponding ownership stroke；4) contour segment that connection generates obtains final vector strokes extraction result.The vectorization method of the present invention can accurately realize the stroke extraction of font, and completely without artificial intervention, and algorithm with automatic running, can be conducive to the batch operation of vector quantization stroke extraction.
Description
Technical field
The present invention relates to a kind of Chinese-character stroke extraction methods of vector font.More particularly, it relate to a kind of non-
Stiff points collection register method, and the result of point of use collection registration proposes a kind of algorithm of extraction Chinese Vector Characters stroke, to complete
The stroke of vector font is automatically extracted, computer graphics and technical field of image processing are belonged to.
Background technology
Chinese character is a kind of ancient word, possesses nearly 5,000 years long histories.In the landslide evolution process of Chinese character
There is diversified font, early stage development of computer, Hanzi font library only includes black matrix, the Song typeface, regular script and imitation Song-Dynasty-style typeface etc.
Several limited fonts.With the development of computer technology and popularizing for network, people more compel the demand of different fonts
It cuts.
Making a set of character library at present, there are mainly two types of methods：One is the writings by being accomplished manually entire character library collection, then
Font is scanned and is handled, in storage to character library；Another kind is the font that each Chinese character is designed by manually, and storage is arrived
In character library.It makes a character library and takes around 2-3 man-years, it is less efficient.So the character library generation method of automation becomes current
The important topic of technology of Chinese character researcher.
Automation generates character library, and there is the property of structuring, i.e. Chinese character pattern to be made of stroke and component based on Chinese character, phase
Same stroke and component can form different Chinese characters by spelling group.It, can be with by carrying out stroke extraction to sub-fraction Chinese character
Entire Hanzi font library is generated using algorithm spelling group.It can be seen that stroke extraction is a key technology that automation generates character library.
Current existing stroke extracting method is all based on the Chinese-character stroke extraction of image.And it is known that font stores
Evolution be to develop to figure font from image font, using Bezier curve come indicate font have higher accuracy and
More flexible storage and processing mode.If existing achievement in research is optimized and combined the font based on figure, should be able to obtain
More accurate ideal experimental result.
Invention content
It, can be effectively it is an object of the invention to propose a kind of Chinese-character stroke extraction method towards vector font
For input vector font library extract each font high quality stroke.Point set is registered as a core of the algorithm simultaneously
Step, for this problem, the present invention proposes the registration of the non-rigid point set based on structural information, to effectively increase point set
The accuracy of registration.
To achieve the goals above, a kind of Chinese-character stroke extraction method towards vector font proposed by the present invention,
It include mainly following four step：
The first step, to target font (below referred to as " data word ", corresponding skeleton point set and profile to be extracted
Point set is referred to as " data point set " and " data profile ") and template database (such as regular script, row pattern) in band corresponding thereto
There are template font (below referred to as " template word ", corresponding skeleton point set and the profile point set abbreviation of stroke categorizing information
" template point set " and " template contours ") carry out skeletal extraction.After the step is finished, two point sets, i.e. data point set are obtained
With template point set.
Second step, using the stroke attaching relation of data point set, template point set and template point set as based on structural information
Non-rigid point set registration algorithm input, the stroke attaching relation of data point set is obtained after registration.That is, herein
After step is finished, the skeleton of target word has been partitioned into several subsets, and each subset includes several skeletal points.
Third walks, and the stroke attaching relation of data point set is converted into the attaching relation of data profile section, which completes
Afterwards, for each contour segment of composition character contour, there is corresponding ownership stroke.But it is right for each stroke institute
The contour segment answered, they are that interruption is inc, this just needs an algorithm to be closed each discrete contour segment set
Operation.
4th step, is closed the contour segment that third step generates, and result is extracted to obtain final stroke.
The above-mentioned first step is specifically described below to the implementation process of the 4th step.
1. the framework extraction method in the first step
It is preferable to use bibliography (TY Zhang and Ching Y.Suen, " A fast parallel for the step
algorithm for thinning digital patterns,”Communications of the ACM,vol.27,
No.3, pp.236-239,1984.) in parallel connectivity keep skeletal extraction algorithm carry out skeletal extraction.It should be noted that
It is that the step can also use other skeletal extraction algorithms, such as bibliography (Arcelli, Carlo, Di Baja, and
Gabriella Sanniti."A width-independent fast thinning algorithm."Pattern
Analysis and Machine Intelligence,IEEE Transactions on 4(1985):463-474.) in
Width independence thinning algorithm and bibliography (Holt, Christopher M., et al."An improved parallel
thinning algorithm."Communications of the ACM 30.2(1987):156-160.) in optimization simultaneously
Row thinning algorithm etc..
2. the non-rigid point set registration algorithm based on structural information in second step
Non-rigid point set registration algorithm based on structural information is to be based on bibliography (Andriy Myronenko and
Xubo Song,“Point set registration:Coherent point drift,”Pattern Analysis and
Machine Intelligence, IEEE Transactions on, vol.32, no.12, pp.2262-2275,2010.) in
A kind of improvement of the point set consistency moving algorithm (Coherent Point Drift, CPD) of proposition, original CPD algorithms exist
It is minimum so that object function obtains after the convergence of expectation maximization (Expectation Maximization, EM) algorithm iteration
It is worth, the state that algorithm reaches after iteration convergence is a kind of state that template point set reaches global optimum.But in fact most
The position of number when point can more be influenced by partial points.Intuitively, for each local point set when iteration convergence,
Its state is not necessarily optimal.
It is added in original CPD algorithms to solve the above-mentioned problems and by structured message, the present invention proposes two aspects
Solution.First, attraction matrix G is optimized, addition localization Optimizing operator (Localized Operator,
LO), it can ensure the influence for being a little more affected by the point inside structural categories in the process of moving in this way, and as much as possible
Ignore the influence of the peripheral point other than classification.Specifically, in original CPD algorithms, matrix G is the matrix of a M*M, the square
Battle array can then be calculated using the following formula：
Wherein yiIndicate that i-th of data point, β are normalized parameter.The attraction matrix after localization Optimizing operator is added
It can be calculated by following formula：
Wherein lijIllustrate the attaching relation of and at i-th point at j-th point, if belong to same subset, l at 2 pointsijIt is 1, it is no
It is then 0.
The Point set matching of localization is added later, thus by the global iterative of original CPD algorithms and office proposed by the present invention
Portion's iteration is combined, and forms a comprehensive point set registration algorithm based on iteration, and the present invention is referred to as SGCPD algorithms.
Based on above-mentioned optimum ideals, algorithm basic flow chart as shown in Figure 1 is obtained.Input is that two point sets are (dark-grey
Color represents data point set, and light gray represents template point set), two point sets of input are carried out with localization Optimizing operator first
Consistency point set moving algorithm (Coherent Point Drift with Localized Operator, CPDLO), the mistake
Journey is the point set registration in global scope.The first of data point set can be obtained after registration according to the information in ownership matrix P
Secondary division result.Then it is directed to each subset and executes the consistency point set moving algorithm partially with localization Optimizing operator
(Localized Coherent Point Drift with Localized Operator, LCPDLO), the process are parts
Point set registration.Global registration is returned after the completion of registration, and cycle is exited until iteration convergence obtains final registering result.
Non-rigid point set is registered, the present invention can also use bibliography (Zhengyou Zhang, " Iterative
point matching for registration of free-form curves and surfaces,”
International journal of computer vision, vol.13, no.2, pp.119-152,1994.) in mention
Iteration point collection registration (ICP) algorithm or bibliography (Haili Chui and Anand Rangarajan, " A new
algorithm for non-rigid point matching,”in Computer Vision and Pattern
Recognition, 2000.Proceedings.IEEE Conference on.IEEE, 2000, vol.2, pp.44-51.) in
The TPS-RPM algorithms mentioned or other commonly use point set registration algorithm to realize.
3. the method that the attaching relation of data point set is converted into the attaching relation of data profile section in third step
For the starting point S and terminating point E of each contour segment, using arest neighbors mode calculate its first ownership S1,
The ownership of E1 and second S2, E2.First ownership refers to the ownership corresponding to the skeletal point nearest apart from current point, and the second ownership is
Refer to the ownership corresponding to the skeletal point different and nearest apart from current point with the first ownership.A contour segment can be obtained in this way
The candidate collection { S1, S2, E1, E2 } of ownership.Then contour segment ownership candidate collection is once traversed, calculates each and returns
Belong to the skeletal point in stroke to the minimum value of two end-point distances sums, is belonged to using the stroke corresponding to the minimum value as current
Finally corresponding stroke belongs to contour segment.
The specific execution flow of the algorithm is as shown in Figure 3.Two end is calculated for the stroke section in dark circle in left figure
Corresponding first and second ownership of point, to which the stroke candidate collection for obtaining to belong to is { stroke a, stroke b, stroke
C }, middle figure indicates the distance to each candidate stroke calculating skeletal point to two endpoints, it can be clearly seen that corresponding to stroke b
Distance is shortest, therefore the final contour segment is attributed to stroke b, as shown on the right.
4. the method for carrying out closed procedure to contour segment set in the 4th step
The contour segment collection corresponding to stroke is defined first and is combined into S, each element represents a discrete profile in set
Section.
The first step constitutes element in stroke Extreme points set E, E for two endpoints of each contour segment in set S
Number should be two times of element number in S.
Second step, the distance of different endpoints between any two, constructs the distance matrix M of endpoint in set of computations E.
Third walks, according to heuristic rule to the distance between point is updated two-by-two in M.Heuristic rule is based primarily upon
Following three points rule：
If a) being located at the same contour segment at 2 points, the distance between they increase 2000-3000, and (numerical value indicates profile
Weight when section connection reduces degree, and the weight connected between bigger 2 points of numerical value is smaller), as shown in Fig. 4 (a), each endpoint
It should be connected with identical point is numbered with it, it can be seen that the point of all pairings is respectively positioned on different contour segments.
If b), 2 lines have passed through skeleton, and 2 points of distance increases 1000-1500, the rule based on the fact that：
Rational stroke section connection is normally at the end or centre position of stroke, and the connection of these positions is usually that will not pass through bone
Frame.As shown in Fig. 4 (b), consider that the connection for the endpoint that circle is irised out, candidate point have 1 and 2, the distance at midpoint 2 to endpoint is more
It is closer, but since the line of 2 to endpoint of point has passed through skeleton, so selected element 1 is as final connection.
If the line c) between 2 points is located at except the effective coverage of font, distance increases 1000-1500, here font
Effective coverage refers to that the region for needing to be painted in rendering glyphs that character contour is surrounded, the rule are based on following things
It is real：The rational general inside that can be all located at font region of stroke section connection.As shown in Fig. 4 (c), the endpoint that circle is irised out is considered
Connection, candidate point has 1 and 2, and more recently, but because the line of point 2 to endpoint is located at word to the distance in same place 2 to endpoint
Except shape effective coverage, therefore selected element 1 is used as final tie point.
Preferably, above-mentioned heuristic rule a), priority relationship b), c) are：a)>B)=c), i.e., it is preferential to meet rule
A), then meet rule b) and c).
4th step connects the nearest corresponding endpoint of distance after heuristic rule adjusts therewith, so for each endpoint
The stroke section corresponding to the two endpoints is fused into a stroke section afterwards, and the endpoint after connection is deleted from set E
Fall, while updating the distance value in M.
5th step is closed it contour segment not closed after the completion of the 4th step by force.
Compared with prior art, the positive effect of the present invention is：
The present invention proposes a kind of stroke extracting method of vector font, and pen is solved from the angle different from existing research
Extraction problem is drawn, vector font, which is extracted, can obtain more accurate stroke extraction as a result, and completely without people
The intervention of work, algorithm with automatic running, can be conducive to the batch operation of vector quantization stroke extraction.On the other hand, the present invention proposes
Non-rigid point set registration based on structural information, is the improvement to existing point set register method and supplement, for Chinese character skeleton
Point set registration can obtain more ideal registering result.Specifically, the present invention carries the stroke of Chinese instrument regular script, imitation Song-Dynasty-style typeface and lishu
Success rate is taken to respectively reach 96.95%, 95.31% and 77.70%.
Description of the drawings
Fig. 1 is the non-rigid point set register flow path figure of the present invention based on structural information；
Fig. 2 is vector font stroke extraction algorithm overview flow chart of the present invention；
Fig. 3 is contour segment ownership algorithm schematic diagram of the present invention；
Fig. 4 is the heuristic rule schematic diagram of contour segment connection of the present invention；
Fig. 5 is skeletal extraction result figure of the present invention；
Fig. 6 is the non-rigid point set registering result figure of the present invention based on structural information；
Fig. 7 is vector font contour segment ownership result schematic diagram of the present invention；
Fig. 8 is contour segment closure result schematic diagram of the present invention；
Fig. 9 is that the Chinese-character stroke of vector font of the present invention automatically extracts result schematic diagram.
Specific implementation mode
In order to make the foregoing objectives, features and advantages of the present invention clearer and more comprehensible, below by specific embodiment and
Attached drawing, the present invention will be further described.
Fig. 1 is the non-rigid point set register flow path figure of the present invention based on structural information, including following components：
1) two point sets of input register with the CPD of localization Optimizing operator, the overall situation is obtained after matching most
Excellent point set correspondence.
2) the preliminary stroke classification result of data point set is obtained according to the point set correspondence obtained in step 1).
3) after step 2) is finished, each stroke has corresponded to a template point set and a set of data points, needle
The CPD registrations that the two set corresponding to each stroke localize are (i.e. partially with the CPD notes of localization Optimizing operator
Volume), to make point set reach the state of local optimum.
4) notice that step 3) destroys the state of point set global optimum, at this time return to step 2) global registration is carried out, it repeats
This process can be obtained final registering result until restraining or reaching defined iterations.
Fig. 2 is the flow chart of the Chinese-character stroke extraction method of the present invention towards vector font, specific steps
It is as follows：
1) system carries out skeletal extraction to the font to be extracted and template font of input, and the results are shown in Figure 5 for extraction.
2) the skeleton point set of the skeleton point set of font to be extracted and template font is implemented based on the non-of structural information
Stiff points collection is registered, and specific steps are as it was noted above, registering result is as shown in Figure 6.
3) for the extraction of the stroke of the skeletal point that is obtained in step 2) as a result, map that on vector font contour segment,
Contour segment classification results after mapping are as shown in Figure 7.
4) each contour segment set obtained in step 3) is closed, to after being closed as a result, such as Fig. 8 institutes
Show.
Fig. 9 is that the Chinese-character stroke of vector font of the present invention automatically extracts result schematic diagram.It can be seen that for multiple
The miscellaneous font present invention can obtain ideal stroke extraction result.
It is understood that example described above is only a part of the embodiment of the present invention, rather than whole
Embodiment.Based on the embodiments of the present invention, those skilled in the art are obtained without making creative work
Every other embodiment, shall fall within the protection scope of the present invention.
Claims (10)
1. a kind of Chinese-character stroke extraction method towards vector font, step include：
1) to the template word with stroke categorizing information in target font to be extracted and template database corresponding thereto
Shape carries out skeletal extraction, obtains data point set and template point set；
2) it using the stroke attaching relation of data point set, template point set and template point set as input, carries out being based on structural information
Non-rigid point set registration, obtain the stroke attaching relation of data point set；
3) the stroke attaching relation of data point set is converted into the attaching relation of data profile section, makes each of composition character contour
A contour segment has corresponding ownership stroke；
4) contour segment that connection generates obtains final vector strokes extraction result.
2. the method as described in claim 1, which is characterized in that step 1) carry out skeletal extraction method be it is following in one
Kind：Parallel connectivity keeps skeletal extraction algorithm, width independence thinning algorithm, optimization parallel thinning algorithm.
3. the method as described in claim 1, which is characterized in that step 2) the non-rigid point based on structural information is focused
Volume, is added to structured message in original point set consistency moving algorithm, by addition localization Optimizing operator, is gone forward side by side
The Point set matching of row localization realizes the point set registration that global iterative and local iteration are combined.
4. method as claimed in claim 3, which is characterized in that in original point set consistency moving algorithm, attract torque
Battle array G is the matrix of a M*M, which is calculated using following formula：
Wherein yiIt indicates that i-th of data point, β are normalized parameter, the attraction matrix after localization Optimizing operator is added by following formula
To calculate：
Wherein lijIndicate the attaching relation of and at i-th point at j-th point, if belong to same subset, l at 2 pointsijIt is 1, is otherwise 0.
5. method as described in claim 3 or 4, which is characterized in that the non-rigid point set based on structural information described in step 2)
Registration includes the following steps：
The consistency point with localization Optimizing operator 2-1) is carried out to the two point set, that is, data point sets and template point set of input
Collect moving algorithm；The process is the point set registration in global scope, according to the information number in ownership matrix after registration
The first division result of strong point collection；
The consistency point set moving algorithm partially with localization Optimizing operator 2-2) is executed to each subset；The process is office
The point set in portion is registered, and registration returns to global registration after the completion, until iteration convergence or reaches defined iterations to get to most
Whole registering result.
6. the method as described in claim 1, which is characterized in that the attaching relation of data point set is converted into data wheel by step 3)
The method of the attaching relation of wide section is：
3-1) to the starting point and ending point of each contour segment, its first ownership and second is calculated using arest neighbors mode
Ownership, the first ownership refer to the ownership corresponding to the skeletal point nearest apart from current point, and the second ownership refers to belonging to not with first
Ownership corresponding to the same and skeletal point nearest apart from current point, to obtain the candidate collection of contour segment ownership；
3-2) candidate collection of contour segment ownership is once traversed, calculates the skeletal point in each ownership stroke to two
The minimum value of end-point distances sum, using the stroke ownership corresponding to the minimum value, as current outline section, finally corresponding stroke is returned
Belong to.
7. the method as described in claim 1, which is characterized in that step 4) is to the method that contour segment is attached：
It 4-1) defining the contour segment collection corresponding to stroke and is combined into S, each element represents a discrete contour segment in set S,
Two endpoints of each contour segment constitute stroke Extreme points set E in set S；
4-2) the distance of different endpoints between any two in set of computations E, constructs the distance matrix M of endpoint；
4-3) according to heuristic rule to the distance between point is updated two-by-two in M；
4-4) for each endpoint, the nearest corresponding endpoint of distance after heuristic rule adjusts therewith is connected, then by this
Stroke section corresponding to two endpoints is fused into a stroke section, and the endpoint after connection is deleted from set E, while more
Distance value in new M；
4-5) for step 4-4) after the completion of not closed contour segment, it is closed by force.
8. the method for claim 7, which is characterized in that step 4-3) heuristic rule includes：
It 4-3a) matches and be attached 2 points is located at different contour segments；
It 4-3b) matches and the line between be attached 2 points does not pass through skeleton；
It 4-3c) matches and the line between be attached 2 points is located inside the effective coverage of font.
9. method as claimed in claim 8, which is characterized in that step 4-3) priority relationships of three heuristic rules
It is：4-3a)>4-3b)=4-3c).
10. method as claimed in claim 8, which is characterized in that step 4-3) it is described to the distance between point carries out two-by-two in M
Update includes：If 2 points are located at the same contour segment, the distance between they increase 2000-3000, make the point of all pairings
It is respectively positioned on different contour segments；If 2 lines have passed through skeleton, 2 points of distance increases 1000-1500 selection lines and does not wear
The endpoint of more skeleton is attached；If the line between 2 points is located at except the effective coverage of font, distance increases 1000-
1500, select the endpoint that line is located inside font effective coverage to be attached.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510303067.7A CN104992143B (en) | 2015-06-04 | 2015-06-04 | A kind of Chinese-character stroke extraction method towards vector font |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510303067.7A CN104992143B (en) | 2015-06-04 | 2015-06-04 | A kind of Chinese-character stroke extraction method towards vector font |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104992143A CN104992143A (en) | 2015-10-21 |
| CN104992143B true CN104992143B (en) | 2018-10-26 |
Family
ID=54303956
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510303067.7A Active CN104992143B (en) | 2015-06-04 | 2015-06-04 | A kind of Chinese-character stroke extraction method towards vector font |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104992143B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107507129B (en) * | 2016-06-14 | 2019-12-13 | 北京大学 | chinese character scaling method |
| CN106384094B (en) * | 2016-09-18 | 2019-07-19 | 北京大学 | A kind of Chinese word library automatic generation method based on writing style modeling |
| CN107092917B (en) * | 2017-03-24 | 2020-06-02 | 北京大学 | Chinese character stroke automatic extraction method based on manifold learning |
| CN107292936B (en) * | 2017-05-18 | 2020-08-11 | 湖南大学 | Chinese character font vectorization method |
| CN109147469B (en) * | 2018-07-09 | 2021-07-30 | 安徽慧视金瞳科技有限公司 | Calligraphy practicing method |
| CN110909563B (en) * | 2018-09-14 | 2023-07-28 | 新方正控股发展有限责任公司 | Method, apparatus, device and computer readable storage medium for extracting text skeleton |
| CN112633428A (en) * | 2019-10-09 | 2021-04-09 | 北京方正手迹数字技术有限公司 | Stroke skeleton information extraction method and device, electronic equipment and storage medium |
| CN112232022B (en) * | 2020-10-15 | 2024-02-27 | 浙江大学 | Method and system for dynamically adjusting parameters and designing Chinese character patterns and layouts |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101447087A (en) * | 2008-11-20 | 2009-06-03 | 北大方正集团有限公司 | Method for stimulating matrix font effect based on outline font and system thereof |
| CN101996309A (en) * | 2009-08-19 | 2011-03-30 | 北大方正集团有限公司 | Method and relevant device for identifying and transforming stroke contour |
| CN102637078A (en) * | 2012-02-27 | 2012-08-15 | 厦门大学 | Method for generating structurally optimized Chinese character patterns |
| CN102682022A (en) * | 2011-03-15 | 2012-09-19 | 高静敏 | Implementation method for Chinese character holographic movable character library and operation of Chinese character holographic movable character library |
| CN102968764A (en) * | 2012-10-26 | 2013-03-13 | 北京航空航天大学 | Chinese character image inpainting method based on strokes |
| CN104392229A (en) * | 2014-10-09 | 2015-03-04 | 南通大学 | Handwriting identification method based on stroke fragment direction distribution characteristics |
-
2015
- 2015-06-04 CN CN201510303067.7A patent/CN104992143B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101447087A (en) * | 2008-11-20 | 2009-06-03 | 北大方正集团有限公司 | Method for stimulating matrix font effect based on outline font and system thereof |
| CN101996309A (en) * | 2009-08-19 | 2011-03-30 | 北大方正集团有限公司 | Method and relevant device for identifying and transforming stroke contour |
| CN102682022A (en) * | 2011-03-15 | 2012-09-19 | 高静敏 | Implementation method for Chinese character holographic movable character library and operation of Chinese character holographic movable character library |
| CN102637078A (en) * | 2012-02-27 | 2012-08-15 | 厦门大学 | Method for generating structurally optimized Chinese character patterns |
| CN102968764A (en) * | 2012-10-26 | 2013-03-13 | 北京航空航天大学 | Chinese character image inpainting method based on strokes |
| CN104392229A (en) * | 2014-10-09 | 2015-03-04 | 南通大学 | Handwriting identification method based on stroke fragment direction distribution characteristics |
Non-Patent Citations (3)
| Title |
|---|
| A fast parallel algorithm for thinning digital patterns;TY ZHAGN 等;《Communications of the ACM》;19840331;第27卷(第3期);236-239 * |
| Point Set Registration:Coherent Point Drift;Andriy Myronenko 等;《IEEE Transactions on Pattern Analysis and Machine Intelligence》;20100318;第32卷(第12期);2262-2275 * |
| 汉字部件的无失真缩放变换方法研究;孙浩等;《计算机应用研究》;20130603;第30卷(第10期);3155-3158 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104992143A (en) | 2015-10-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104992143B (en) | A kind of Chinese-character stroke extraction method towards vector font | |
| CN110033007A (en) | Attribute recognition approach is worn clothes based on the pedestrian of depth attitude prediction and multiple features fusion | |
| CN113239131B (en) | Low-sample knowledge graph completion method based on meta-learning | |
| CN105046694A (en) | Quick point cloud registration method based on curved surface fitting coefficient features | |
| CN109740537B (en) | Method and system for accurately marking attributes of pedestrian images in crowd video images | |
| CN108304357A (en) | A kind of Chinese word library automatic generation method based on font manifold | |
| CN105243139A (en) | Deep learning based three-dimensional model retrieval method and retrieval device thereof | |
| CN105931267B (en) | A kind of moving object segmentation tracking based on improvement ViBe algorithm | |
| CN106530341A (en) | Point registration algorithm capable of keeping local topology invariance | |
| CN107316327B (en) | Fractured bone model registration method | |
| CN106250909A (en) | A kind of based on the image classification method improving visual word bag model | |
| CN110910433A (en) | Point cloud matching method based on deep learning | |
| Dai et al. | Multiscale residual convolution neural network and sector descriptor-based road detection method | |
| CN104992454A (en) | Regionalized automatic-cluster-change image segmentation method | |
| Liu et al. | A novel rock-mass point cloud registration method based on feature line extraction and feature point matching | |
| CN115116076A (en) | Character recognition method based on stroke extraction | |
| CN106504219A (en) | Constrained path morphology high-resolution remote sensing image road Enhancement Method | |
| He et al. | Nas-lid: Efficient neural architecture search with local intrinsic dimension | |
| CN111079208B (en) | Particle swarm algorithm-based CAD model surface corresponding relation identification method | |
| Ye et al. | FMAM-Net: fusion multi-scale attention mechanism network for building segmentation in remote sensing images | |
| CN115100406B (en) | Weight information entropy fuzzy C-means clustering method based on superpixel processing | |
| CN109636818A (en) | A kind of Laplce's canonical constrains the Target Segmentation method of lower low-rank sparse optimization | |
| Wang et al. | Weakly-supervised salient object detection through object segmentation guided by scribble annotations | |
| CN109255794A (en) | A kind of full convolution edge feature detection method of standard component depth | |
| Zhao et al. | Improved hierarchical clustering with non-locally enhanced features for unsupervised person re-identification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |