CN100367298C - Universal feature describing method for character recognition - Google Patents
Universal feature describing method for character recognition Download PDFInfo
- Publication number
- CN100367298C CN100367298C CNB2004100167330A CN200410016733A CN100367298C CN 100367298 C CN100367298 C CN 100367298C CN B2004100167330 A CNB2004100167330 A CN B2004100167330A CN 200410016733 A CN200410016733 A CN 200410016733A CN 100367298 C CN100367298 C CN 100367298C
- Authority
- CN
- China
- Prior art keywords
- point
- histogram
- value
- interval
- symbol
- 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.)
- Expired - Fee Related
Links
Images
Landscapes
- Character Discrimination (AREA)
- Character Input (AREA)
Abstract
The present invention relates to a universal symbol identification character description method, which has the characteristics of expandability, resistance to noise, deformation and interference, and resistance to rotation and extension. The present invention can be used for identifying handwriting symbols and characters in different kinds of drawing paper and documents. The present invention is characterized in that 1) symbols are divided into points, geometrical constraints between points are used as the base elements of symbol shape description; 2) When any point is selected as a reference point, a corresponding histogram is obtained by accounting the geometrical constraint of every two points among the other points, and each point can be used as the reference point to respectively obtain one histogram corresponding to each point; 3) values in each interval of the histograms corresponding to the all points are synthesized at static meaning or mathematic meaning to construct corresponding character description. Compared with the other method, in 71 items of performance tests, the present invention has 68 best items, 3 items are in second level, and only one item of identification rates is below to 90%.
Description
Technical field
The invention belongs to pattern-recognition and field of artificial intelligence, it specifically is a kind of general character description method that is used for Symbol recognition, can be used for symbol and character recognition in various drawings and the document and the pen input, for example: the Symbol recognition of every field such as circuit diagram, engineering drawing, map, architectural drawing, music score, trade mark, mathematical formulae, optical character, an input character.
Background technology
Symbol recognition is one of important research contents in pattern-recognition, Figure recognition, document analysis and identification field, is playing the part of important role aspect the automatic understanding of circuit diagram (document that sees reference [1] [2]), engineering drawing (document that sees reference [3] [4]), map (document that sees reference [5] [6]), architectural drawing (document that sees reference [7-9]), music score (document that sees reference [10] [11]), trade mark (document that sees reference [12] [13]), mathematical formulae (document that sees reference [14] [15]), optical character (document that sees reference [16] [17]) and the identification.The present research in this area has obtained certain achievement, still, and for the research and development of utility system or far from being enough.Can the difficult point that face at present be: find a method that all shows excellence in versatility, expandability, antinoise and various aspects such as distortion interference performance, rotation and flexible unchangeability, has outstanding combination property.Find a method of having outstanding performance in one aspect not difficult comparatively speaking, all show method excellent, that combination property is outstanding in all fields and be not easy but will find, up to the present, relevant research is also in process.
The performance of a symbol recognition system depends on the character description method that adopts to a great extent.According to the describing method that is adopted, existing Symbol Recognition can be divided into two big classes: structural approach and statistical method.Some representational methods in this two class will be discussed below.
About structural approach, the feature description of symbol roughly all has following characteristics: be some basic geometric elements with symbol decompose at first, these basic geometric elements and the mutual relationship between them constitute the description of a symbol.This class description method can connect with graph model very naturally, and a large amount of Symbol Recognition is to utilize graph model to carry out feature description (document that sees reference [1] [9] [16]).Attributed relational graph (ARG) is the most classical a kind of graph model, and in list of references [16], relative position between the key point and connectedness are used to describe printed Chinese character as attribute.Proposed a kind of new model that is called regional adjacent map (RAG) in the list of references [9], here, the node of figure is represented a polygon, and the arc of figure represents whether there is public boundary between two polygons.Also reported case study in the document [9] based on the building Symbol recognition of regional adjacent map.Another kind of structural approach is based on the geometrical constraint (document that sees reference [4] [7]) between the fundamental element.In list of references [4], be basic geometric element at first with symbol decompose, as: straight line, arc, ring etc., then the geometrical constraint between the fundamental element of given symbol is carried out some hypothesis, again hypothesis is tested, see whether it satisfies corresponding to the predefined one group of rule of each ideal symbol, thereby identify symbol.List of references [4] has provided the result of study about Symbol recognition in the engineering drawing.Other structural approach comprises that also utilization mates (document that sees reference [8]) based on the deformable template and the incoming symbol of line segment.More than the shortcoming of these methods be: system performance depends on picture quality to a great extent.The something in common of these methods is: will be basic geometric element with symbol decompose at first.When noise and distortion appeared in the image, basic geometric element often can not be detected exactly; Equally, the also very difficult relation that obtains exactly between the fundamental element.This means that the correctness that symbolic feature is described can't guarantee, thereby can cause the decline of recognition correct rate.
Different with architectural feature, statistical nature does not need basic geometric element such as straight line, arc, ring, key point is detected, and statistical nature is based on a little basically.For bianry image, point is the most only a fundamental element (document that sees reference [18]), can be directly inputted to sorter and be used for Symbol recognition; But such sorting technique is difficult to guarantee rotation, flexible, translation invariance.The most classical have rotation, statistical nature flexible, translation invariance is invariant moments (document that sees reference [19]); But the information that invariant moments provided is very limited, and the extendible ability of system can't be guaranteed, especially the identification that large character set is closed.The ring projection is the statistical nature (document that sees reference [17]) that another kind has the invariable rotary ability, its computing method are as follows: at first selected center, outwards make concentric circles cutting character with this center, the point of forming character drops on number on each circle as the description to character shape; The shortcoming of the method is: when there is various distortion in character, be difficult to find a stable center of circle, therefore, the method is not suitable for the identification of the hand-written character that contains distortion.Shape Context is a kind of shape description method (document that sees reference [20]) that proposes recently, it be calculated as follows: (1) is the center with each point, makes a histogram, and how the point of adding up other distributes on every side at this point.(2) finish the calculating of form fit degree by the corresponding relation of searching for each point between two shapes.Shape Context is a kind of shape description method with fine antinoise and distortion interference performance, and its rotational invariance obtains by the following method: calculating when being the histogram at center with the every bit, with the tangent line of this point as the x axle.But, such computing method can not resemble and guarantee rotational invariance reliably desired, the reasons are as follows: under noise, often can not obtain definite border, and boundary shape tends to distortion, obtains the stable tangent line at each point can not resembling under ideal state easily.
Above-described the whole bag of tricks has advantage separately, weakness is separately also arranged, up to the present, also the neither one method can possess outstanding combination property, in versatility, antinoise and distortion interference, rotation and various aspects such as flexible unchangeability, expandability gratifying performance is arranged all that is:.
List of references
[1]Groen，F.，Sanderson，A.，Schlag，F.：Symbol recognition in electrical diagrams usingprobabilistic graph matching.Pattern Recognition Letters 3(1985)343-350
[2]Okazaki A.，Kondo T.，Mori K.，Tsunekawa S.，and Kawamoto E，An automatic circuitdiagram reader with loop-structure-based symbol recognition，IEEE T-PAMI 10(1988)331-341
[3]Filipski，A.J.，Flandrena，R.：Automated conversion of engineering drawings to CAD form.Proceedings ofthe IEEE 80(1992)1195-1209
[4]Luo，Y.，Liu，W.，Y.：Engineering drawings recognition using a case-based approach.In：International Conference on Document Analysis and Recognition 2003，Edinburgh，UK
[5]Boatto，L.，Consorti，V.，Del Buono，M.，Di Zenzo，S.，Eramo，V.，Espossito，A.，Melcarne，F.，Meucci，M.，Morelli，A.，Mosciatti，M.，Scarei，S.，Tucci，M.：An interpretation system for landregister maps.Computer 25(1992)25-33.
[6]Samet，H.，Soffer，A.：Marco：Map retrieval by content.IEEE T-PAMI 18(1996)783-797
[7]Ah-Soon，C.，Tombre，K.：Architectural symbol recognition using a network of constraints.Pattern Recognition Letters 22(2001)231-248
[8]Valveny E.，Marti E.：A model for image generation and symbol recognition through thedeformation of linear shapes.Pattern Recognition Letters 24(2003)2857-2867
[9]Llados J.，Marti E.，Villanueva J.J.：Symbol recognition by error-tolerant subgraph matchingbetween region adjacency graphs.IEEET-PAMI 23(2001)1137-1143
[10]Yadid-Pecht，O.，Gerner，M.，Dvir，L.，Brutman，E.，Shimony，U.：Recognition ofhandwritten musical notes by a modified neocognitron.Machine Vision and Applications 9(1996)65-72
[11]Rossant F.：A global method for music symbol recognition in typeset music sheets.PatternRecognition Letters 23(2002)1129-1141
[12]Chang，M.，Chen，S.：Deformed trademark retrieval based on 2D pseudo-hidden Markovmodel.Pattern Recognition 34(2001)953-967
[13]Cortelazzo，G.，Mian，G.，Vezzi，G.，Zamperoni，P.：Trademark shapes description by stringmatching techniques.Pattern Recognition 27(1994)1005-1018
[14]Lee，H.J.，Lee，M.C.：Understanding mathematical expression in a printed document.Proceedings of the 2nd International Conference on Document Analysis and Recognition，1993，502-505
[15]Chaudhuri，B.B.，Garain，U.：An approach for recognition and interpretation ofmathematical expressions in printed document.Pattern analysis and applications 3(2000)120-131
[16]Huang X.，Gu J.，Wu Y.：A constraint approach to multifont Chinese character recognition.IEEET-PAMI 15(1993)838-843
[17]Yuen P. C.，Feng G.C.，Tang Y. Y.：Printed Chinese character similarity measurement usingring projection and distance transformation.International Journal of Pattern Recognition andArtificial Intelligence 12(1998)209-221
[18]Schurmann J.：Pattern classification，a unified view of statistical and neural approach.JohnWiley&Sons(New York)1996
[19]Hu M.K.：Visual pattern recognition by moment invariants.IRE Transaction onInformation Theory 8(1962)179-187
[20]Belogie S.，Malik J.，Puzicha J.：Shape matching and object recognition using shapecontexts.IEEE T-PAMI 24(2002)509-520
[21]Chen，K.Z.，Zhang，X.W.，Ou，Z.Y.，Feng，X.A.：Recognition of digital curves scannedfrom paper drawings using genetic algorithm，Pattern Recognition 36(2003)123-130
[22]Electronic Proceedings of 5th IAPR International Workshop on Graphics Recognition(GREC 2003)，2003，Barcelona，Spain
[23]
www.cvc.uab.es/grec2003。
Summary of the invention
The objective of the invention is to overcome the deficiencies in the prior art part, a kind of outstanding combination property, general, as to can be used for Symbol recognition character description method (building method of proper vector) that possesses is provided.
The general character description method that is used for Symbol recognition (building method of proper vector) that the present invention proposes, concrete steps are as follows:
(a) be a little with symbol segmentation, with the fundamental element of the geometrical constraint between point and the point as the symbol shape description; The definition of the geometrical constraint between point and point correspondingly will obtain different feature descriptions not simultaneously;
(b) when any one point is selected as reference point (initial point), other each point geometrical constraint between any two added up to obtain a corresponding histogram; Put as with reference to point with each respectively, then obtain a histogram respectively corresponding to each point;
(c) will carry out comprehensively describing (proper vector) on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct characteristic of correspondence, when adopting different computing method each histogram is carried out can correspondingly obtaining different feature descriptions when comprehensive;
(d) two or more different feature descriptions are combined, can constitute a new feature description.
Among the present invention, the geometrical constraint between described point and the point can be defined as follows: with these 2 and reference point is that the summit forms a triangle, with vertex of a triangle, the length of side, angle be independent variable any type of function and with its mathematics on amount of equal value.
Among the present invention, the definition of geometrical constraint between described point and the point, can be these two points respectively and reference point connect angle between the line that is constituted.
Among the present invention, the definition of geometrical constraint between described point and the point, can also be these two points respectively and reference point connect length ratio between straight-line segment short in the resulting line and the long straight-line segment.
Among the present invention, describedly will carry out comprehensively describing on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct characteristic of correspondence, its computing method can be: each histogram is dropped on each interval interior value add up, can obtain a new histogram corresponding to each interval; The histogrammic value that all these are new and the value of any type of function thereof constitute a characteristic of correspondence and describe.
Among the present invention, describedly will carry out comprehensively describing on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct characteristic of correspondence, its computing method also can be: each histogram is dropped on each interval interior value add up, average, any type of function of N rank central moment, N rank moment of the orign and above-mentioned statistic, N is a real number; This tittle that try to achieve in all intervals constitutes a feature description.
Among the present invention, describedly will carry out comprehensively describing on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct characteristic of correspondence, its computing method can also be: each histogram is dropped on sue for peace after each interval interior value is asked Nth power earlier again, N is a real number, for example, 2≤N≤1000; That try to achieve in each interval and the value of any type of function constitute a feature description.
That the Symbol recognition character description method that the present invention proposes has is extendible, antinoise and distortion is disturbed, rotation and flexible constant characteristics, can be used for symbol and character recognition in various drawings and the document and the pen input.
Description of drawings
Fig. 1: with P
0The synoptic diagram of geometrical constraint between other 2 during for reference point
Embodiment
A symbol recognition system is made up of following steps usually: pre-service, feature extraction, classification.Here, pre-service adopts the method that proposes in the list of references [21] that symbol carry out thinization; Sorter adopts nearest neighbor method (the simplest a kind of sorter); The method that feature extraction adopts the present invention to propose, embodiment is as follows:
Embodiment 1:
(1) supposes that the symbol skeleton that extracted by pre-service is made of note work: P N point
0, P
1..., P
N-1Respectively with P
0, P
1..., P
N-1As the reference point, ask other each point geometrical constraint between any two.Here, between 2 geometrical constraint be defined as these two points respectively and reference point connect angle between the line constituted.For example shown in Figure 1, with P
0Be reference point, P
iAnd P
jBetween geometrical constraint be ∠ P
iP
0P
j, here, P
i∈ { P
1, P
2..., P
N-1And P
j∈ { P
1, P
2..., P
N-1.
(2) with P
kDuring for reference point, with P
iAnd P
jBetween geometrical constraint note make C
Ij(P
k), to { C
Ij(P
k) | i=0,1 ..., N-2; J=i+1, i+2 ..., N-1} adds up, and can obtain a histogram, and note is done: H (P
k), here, i ≠ k and j ≠ k.Respectively with P
0, P
1.., P
N-1As the reference point, can obtain N corresponding histogram H (P
0), H (P
1) ..., H (P
N-1).
(3) each histogram can be expressed as a vector, supposes that each histogram respectively has M interval, histogram H (P
k) vector form note make H (P
k)=[H
1(P
k), H
2(P
k) ..., H
M(P
k)], k=0,1 ..., N-1; Here, H
j(P
k) expression histogram H (P
k) in j interval value; To each histogrammic i interval value { H
i(P
0), H (P
1) ..., H
i(P
N-1) add up, can obtain a new histogram, note is made F (i); In the manner described above, can obtain a new histogram corresponding to each interval, can obtain altogether M histogram F (i) | i=1,2 ..., M}; Suppose that histogram F (i) has L interval, its vector form note is made F (i)=[F
1(i), F
2(i) ... F
L(i)], i=1,2 ..., L; Here, F
j(i) j interval value of expression histogram F (i); [F
j(i) | i=1,2 ..., M; J=1,2 ..., L] be the proper vector of being asked that is used for denotational description.
Embodiment 2:
Compare with embodiment 1, the definition of geometrical constraint, remainder is identical between 2 o'clock.Here, between 2 geometrical constraint be defined as these two points respectively and reference point connect in the resulting line length ratio between short straight-line segment and the long straight-line segment.For example shown in Figure 1, with P
0Be reference point, P
iAnd P
jBetween geometrical constraint be min{|P
0P
i|/| P
0P
j|, | P
0P
j|/| P
0P
i|, here, | P
0P
i| and | P
0P
j| represent line segment P respectively
0P
iAnd P
0P
jLength.
Embodiment 3:
Embodiment 1 and embodiment 2 respectively generate an independent feature vector, and note is made [F respectively
j(i) | i=1,2 ..., M; J=1,2 ..., L] and [G
j(i) | i=1,2 ..., M '; J=1,2 .., L '], these two vectors are joined end to end line up a vector and be the proper vector that embodiment 3 is asked.
Based on the character description method of embodiment 3, the inventor has designed corresponding Symbol recognition program.After tested, multinomial test correct recognition rata reaches 100%, has only one to be lower than 90%, is 86.4%.In all 71 tests, compare with other method, 68 performances are best, comprising: all antinoises, distortion, noise add the test of distortion, rotation, and part is anti-flexible, the flexible test that adds rotation etc.Method of testing is the input piece image, finds the model the most similar to it (ideal image).6850 width of cloth images have been tested altogether.DCO the results are shown in Table 1～5.Average recognition time is less than 1 second for each symbol.
Table 1: the discrimination of ideal image
| Pattern number | 50 | |
| Symbolic number | 5 | 20 |
| Picture number | 5 | 20 |
| Discrimination (%) | 100 | 100 |
Table 2: the discrimination of the flexible image of rotation and size
| Pattern number | 50 | ||
| Symbolic number | 5 | 20 | 50 |
| Picture number | 25 | 100 | 250 |
| Rotation (%) | 100 | 99 | 97.2 |
| Flexible (%) | 100 | 98 | 96.4 |
| Rotation adds flexible (%) | 100 | 92 | 86.4 |
Table 3: the discrimination of deformation pattern
| Pattern number | 50 | |
| Symbolic number | 5 | 15 |
| Picture number | 25 | 75 |
| Rank 1 (%) | 100 | 100 |
| Rank 2 (%) | 100 | 100 |
| Rank 3 (%) | 100 | 98.6 |
Table 4: the discrimination that contains noise image
| Pattern number | 5 | 20 | 50 |
| Symbolic number | 5 | 20 | 50 |
| Picture number | 25 | 100 | 250 |
| Rank 1 (%) | 100 | 100 | 100 |
| Rank 2 (%) | 100 | 99 | 100 |
| Rank 3 (%) | 100 | 100 | 99.6 |
| Rank 4 (%) | 100 | 100 | 97.2 |
| Rank 5 (%) | 100 | 100 | 93.2 |
| Rank 6 (%) | 100 | 100 | 96.4 |
| Rank 7 (%) | 100 | 94 | 95.6 |
| Rank 8 (%) | 100 | 99 | 90 |
| Rank 9 (%) | 100 | 99 | 91.2 |
Table 5: noise adds the discrimination (pattern number: 15 of deformation pattern; Symbolic number: 15; Picture number: 75)
| Noise | Distortion | ||
| Rank 1 (%) | Rank 2 (%) | Rank 3 (%) | |
| Rank 1 (%) | 100 | 100 | 97.3 |
| Rank 2 (%) | 100 | 100 | 98.7 |
| Rank 3 (%) | 100 | 100 | 96 |
| Rank 4 (%) | 100 | 100 | 98.7 |
| Rank 5 (%) | 98.7 | 94.7 | 93.3 |
| Rank 6 (%) | 100 | 100 | 94.7 |
| Rank 7 (%) | 100 | 100 | 98.7 |
| Rank 8 (%) | 100 | 100 | 100 |
| Rank 9 (%) | 100 | 100 | 100 |
Claims (7)
1. general character description method that is used for Symbol recognition, it is characterized in that: (1) is a little with symbol segmentation, the fundamental element of describing as symbol shape with the geometrical constraint between point and the point; (2) optional some point for referencial use is added up other each point geometrical constraint between any two, obtains a corresponding histogram; Put as with reference to point with each respectively, then obtain a histogram respectively corresponding to each point; (3) will carry out comprehensively describing on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct a characteristic of correspondence; (4) two or more different feature descriptions are combined, obtain a new feature description.
2. method according to claim 1, it is characterized in that: the geometrical constraint between described point and the point is defined as follows: with these 2 and reference point is that the summit forms a triangle, with vertex of a triangle, the length of side, angle be independent variable any type of function and with its mathematics on amount of equal value.
3. method according to claim 1 and 2 is characterized in that: the geometrical constraint between described point and the point is defined as follows: these two points respectively and reference point connect angle between the line constituted.
4. method according to claim 1 and 2 is characterized in that: the geometrical constraint between described point and the point is defined as follows: these two points respectively and reference point connect in the resulting line length ratio between short straight-line segment and the long straight-line segment.
5. method according to claim 1, it is characterized in that: describedly will carry out comprehensively describing on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct characteristic of correspondence, its computing method are as follows: each histogram is dropped on each interval interior value add up, can obtain a new histogram corresponding to each interval; The histogrammic value that all these are new and the value of any type of function thereof constitute a characteristic of correspondence and describe.
6. method according to claim 1, it is characterized in that: describedly will carry out comprehensively describing on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct characteristic of correspondence, its computing method are as follows: each histogram is dropped on each interval interior value add up, average, any type of function of N rank central moment, N rank moment of the orign and above-mentioned statistic, N is a real number; This tittle that try to achieve in all intervals constitutes a feature description.
7. method according to claim 1, it is characterized in that: describedly will carry out comprehensively describing on statistical significance or the mathematical meaning corresponding to histogrammic each interval value of being had a few to construct characteristic of correspondence, its computing method are as follows: each histogram is dropped on sue for peace after each interval interior value is asked Nth power earlier again, N is a real number; That try to achieve in each interval and the value of any type of function constitute a feature description.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100167330A CN100367298C (en) | 2004-03-04 | 2004-03-04 | Universal feature describing method for character recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100167330A CN100367298C (en) | 2004-03-04 | 2004-03-04 | Universal feature describing method for character recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1560790A CN1560790A (en) | 2005-01-05 |
| CN100367298C true CN100367298C (en) | 2008-02-06 |
Family
ID=34440618
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2004100167330A Expired - Fee Related CN100367298C (en) | 2004-03-04 | 2004-03-04 | Universal feature describing method for character recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100367298C (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101630335B (en) * | 2008-07-18 | 2012-01-25 | 纬衡浩建科技（深圳）有限公司 | Method for comparing similarity and difference between drawings |
| DE102011075275A1 (en) * | 2011-05-04 | 2012-11-08 | Bundesdruckerei Gmbh | Method and device for recognizing a character |
| CN103258200B (en) * | 2013-03-07 | 2016-08-24 | 大连理工大学 | A kind of Symbol Recognition based on affine invarient |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000022899A (en) * | 1998-07-06 | 2000-01-21 | Minolta Co Ltd | Image recognition device |
| CN1252585A (en) * | 1999-11-08 | 2000-05-10 | 广东省科学院自动化工程研制中心 | Local prefering matching method of limited sign set |
-
2004
- 2004-03-04 CN CNB2004100167330A patent/CN100367298C/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000022899A (en) * | 1998-07-06 | 2000-01-21 | Minolta Co Ltd | Image recognition device |
| CN1252585A (en) * | 1999-11-08 | 2000-05-10 | 广东省科学院自动化工程研制中心 | Local prefering matching method of limited sign set |
Non-Patent Citations (3)
| Title |
|---|
| 手写数学符号的基元识别方法. 赵学军等.重庆大学学报，第21卷第2期. 1998 * |
| 数字地图符号的形状描述与识别. 黄文骞.测绘学报，第28卷第3期. 1999 * |
| 票据自动识别录入系统. 杨颖，杨磊.计算机工程与应用. 2002 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1560790A (en) | 2005-01-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Moreno-García et al. | New trends on digitisation of complex engineering drawings | |
| Xie et al. | Robust clustering by detecting density peaks and assigning points based on fuzzy weighted K-nearest neighbors | |
| Heutte et al. | A structural/statistical feature based vector for handwritten character recognition | |
| Lladós et al. | Symbol recognition: Current advances and perspectives | |
| Rebelo et al. | Optical recognition of music symbols: A comparative study | |
| Mahmoud | Recognition of writer-independent off-line handwritten Arabic (Indian) numerals using hidden Markov models | |
| Lee et al. | An efficient graph-based recognizer for hand-drawn symbols | |
| Sahare et al. | Script identification algorithms: a survey | |
| Santosh | Document image analysis: Current trends and challenges in graphics recognition | |
| Jouili et al. | Graph matching based on node signatures | |
| CN100367298C (en) | Universal feature describing method for character recognition | |
| Montagner et al. | Staff removal using image operator learning | |
| Sidere et al. | Vector representation of graphs: Application to the classification of symbols and letters | |
| Tombre et al. | Pattern recognition methods for querying and browsing technical documentation | |
| Tümen et al. | DPFrag: trainable stroke fragmentation based on dynamic programming | |
| Rusinol et al. | CVC-UAB's Participation in the Flowchart Recognition Task of CLEF-IP 2012. | |
| Zhu et al. | Segmentation of on-line freely written Japanese text using SVM for improving text recognition | |
| Qureshi et al. | Combination of symbolic and statistical features for symbols recognition | |
| Darvishzadeh et al. | Finding the Answer: Techniques for Locating Students' Answers in Handwritten Problem Solutions | |
| Datta et al. | Detection and identification of logic gates from document images using mathematical morphology | |
| Sánchez et al. | A platform to extract knowledge from graphic documents. application to an architectural sketch understanding scenario | |
| Nawade et al. | Old handwritten music symbol recognition using directional multi-resolution spatial features | |
| Prasanthkumar et al. | Word level Script and Language identification for Unconstrained handwritten document images | |
| Liang et al. | Sketch matching on topology product graph | |
| Ribas et al. | Fractal dimension of bag-of-visual words |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20080206 Termination date: 20180304 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |