CN101038627A - Method for recognizing print hand Arabic alphabets based on boundary characteristic - Google Patents
Method for recognizing print hand Arabic alphabets based on boundary characteristic Download PDFInfo
- Publication number
- CN101038627A CN101038627A CN 200710072163 CN200710072163A CN101038627A CN 101038627 A CN101038627 A CN 101038627A CN 200710072163 CN200710072163 CN 200710072163 CN 200710072163 A CN200710072163 A CN 200710072163A CN 101038627 A CN101038627 A CN 101038627A
- Authority
- CN
- China
- Prior art keywords
- straight line
- length
- boundary
- border
- ripple
- 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
Images
Abstract
The invention provides a print-form Arabic alphabet identification method based on boundary characteristic, wherein the upside, downside, left and right boundaries is regarded as a wave, each boundary is regarded as an assembly of a series of wave elements, and the following boundary characteristics are extracted therefrom: the number of the component wave, the number of the zero line, the length of the first zero line in the right boundary, the length of the first zero line in the downside boundary, the length of the longest zero line in the upside boundary, the length of the longest zero line in the right boundary, the length of the longest zero line in the upside boundary, the length of the longest zero line in the downside boundary and the number of the positive line in the upside boundary, moreover, the height width ratio of the alphabets and the alphabet auxiliary part are involved as the identification characteristics, finally, each print-form Arabic alphabet is identified by using four decision trees according to the four kinds of formats of the alphabets-the independent, the beginning, the middle and the end.
Description
(1) technical field
The present invention relates to a kind of optical character recognition method, particularly relate to a kind of method for recognizing print hand Arabic alphabets.
(2) background technology
In various spoken and written languages, Arabic is one of most widely used literal.Comprise alphabetical Lam-Alif, Arabic alphabet has 29.In simple terms, Arabic alphabet has following characteristics:
Each Arabic alphabet according to the diverse location in word have 2～4 kinds multi-form, be respectively independence, beginning, centre and ending form;
2. Arabic does not have vowel, and vowel is to embody by the method that is called the additional character of " moving symbol " to the consonant mark." the moving symbol " of two expression vowels arranged in Arabic, be respectively Hamaza and Madda;
3. alphabetical Lam and alphabetical Alif can connect together and form a new alphabetical Lam-Alif;
4. pitch is unequal, and what not only difference was alphabetical may be not wide, and four kinds of forms of certain letter are not wide yet;
5. most of Arabic alphabet contains slave part, and these slave parts may be positioned at the top, following of alphabetical main part or be positioned at letter inner;
6. adjacent several Arabic alphabets may have identical main body form in alphabet, and are only distinguished with slave part, and this feasible identification for slave part becomes very important.
Research for Arabic identification can be traced back to 1975 the earliest, and Nazif has realized a kind of blocks letter Arabic literary composition recognition system in its master thesis.After this research for Arabic never has any progress, up to 1980, has risen the upsurge of one Arabic Study of recognition, and this upsurge continues up to the beginning of the nineties.In this period, people such as Amin, Shoukry, Jambi are to the big quantity research of Arabic identification having carried out.At present, the method for recognizing print hand Arabic alphabets that proposes of researchers mainly comprises: based on the recognition methods of image density; Recognition methods based on square invariant features and neural network; Recognition methods based on primitive feature and decision tree.In the middle of these methods, though based on image density and fairly simple based on the recognition methods of square invariant features, algorithm speed is slow, and these methods do not make full use of the abundant resemblance of Arabic alphabet; Recognition methods based on primitive feature needs thinning process, and its recognition speed is slow, because the Arabic alphabet complex structure phenomenons such as fracture, burr can occur in thinning process, causes the discrimination of this method low simultaneously.For print hand Arabic alphabets identification, be badly in need of a kind of simple, high, fireballing method of discrimination, for the performance that improves existing commercial Arabic automatic recognition system is laid a good foundation.
(3) summary of the invention
The objective of the invention is to propose a kind of method for recognizing print hand Arabic alphabets based on boundary characteristic.
The technical solution adopted in the present invention is the first step, four borders, upper and lower, left and right of calculating letter; Second step, regard each border as a ripple, with each boundary representation the set of a series of ripple unit; In the 3rd step, from the set of ripple unit, extract ripple unit number, zero straight line number, the longest zero straight length as recognition feature; At last, adopt decision tree to discern each print hand Arabic alphabets.In order to improve algorithm speed, adopt from top to bottom, from top to bottom, from left to right, the mode of four direction scanning from right to left four borders of calculating letter.This border computing method are not only simple, and computing velocity will be greatly faster than traditional edge following algorithm.In order to simplify identifying, be the set of zero straight line, positive straight line and three kinds of ripple units of negative straight line with each bar boundary representation, and to its merge, deletion action to be to improve the robustness of algorithm.In feature extraction phases, extract ripple unit number, zero straight line number, zero straight length, alphabetical depth-width ratio etc. as recognition feature, these features are not only simple, and can directly obtain from border set or letter image very soon.Decision tree at first utilizes alphabetical depth-width ratio that print hand Arabic alphabets is divided three classes, for each class, utilize following feature again: in the zero straight line number in ripple in the depth-width ratio of alphabetical slave part, each border unit number, each border, the right margin in the length of article one zero straight line, the lower boundary in the length of article one zero straight line, the coboundary in the length of the longest zero straight line, the right margin in the length of the longest zero straight line, the lower boundary in the length of the longest zero straight line and the coboundary number of positive straight line discern print hand Arabic alphabets.
The present invention has overcome the shortcoming that existing method for recognizing print hand Arabic alphabets speed is slow, algorithm is complicated, its beneficial effect is, be the set of various ripple unit with alphabetical boundary representation and therefrom extract the feature of various ripple unit, characteristic extraction procedure is simple and clear, and algorithm speed is fast, be easy to realize.
(4) description of drawings
Fig. 1 is the decision tree of the print hand Arabic alphabets of identification absolute version;
Fig. 2 is the decision tree of the print hand Arabic alphabets of identification beginning form;
Fig. 3 is the decision tree of the print hand Arabic alphabets of identification intermediate form;
Fig. 4 is the decision tree of the print hand Arabic alphabets of identification ending form.
(5) embodiment
The invention will be further described below in conjunction with the drawings and specific embodiments:
In conjunction with Fig. 1-Fig. 4, wherein, HW is alphabetical depth-width ratio, and W is a pitch, and H is the letter height, LN
1Be the unit of the ripple in left margin number, LN
2Be the unit of the ripple in coboundary number, LN
3Be the unit of the ripple in right margin number, LN
4Be the unit of the ripple in lower boundary number, SN
1Be the zero straight line number in the left margin, SN
2Be the zero straight line number in the coboundary, SN
3Be the zero straight line number in the right margin, SN
4Be the zero straight line number in the lower boundary, SL
31Be the length of zero straight line of article one in the right margin, SL
41Be the length of zero straight line of article one in the lower boundary, MSL
2Be the length of the longest zero straight line in the coboundary, MSL
3Be the length of the longest zero straight line in the right margin, MSL
4Be the length of the longest zero straight line in the lower boundary, PN
2Be the number of positive straight line in the coboundary, SHW is the depth-width ratio of slave part.
In order to extract boundary characteristic, at first from letter image, calculate four borders of letter in the present embodiment.The computing method following (is example with the left margin) on letter border: from left to right scan letter image, establish e
pThe number of the continuous white pixel when running into first black pixel for scanning from letter image p capable left side beginning to the right, then Zi Mu left margin E can be expressed as: E={e
p| p=1,2 ...., N}, wherein N is the letter height.The computing method on its excess-three border are similar, just the direction of scanning difference.The direction of scanning of coboundary is for from top to bottom, and right margin is for from right to left, and lower boundary is for from top to bottom.
Then, utilize following steps that each boundary representation is ripple unit's set (is example with the left margin).
De
p=de
P-1If de
P-1=de
P+1(2)
Rule 1: if exist p and q to satisfy p＜q, de
P-1≠ 0, de
Q+1≠ 0, and de
p=de
P+1=... .=de
Q-1=0, then the border between p and the q is one zero straight line, and its length is q-p.
Rule 2: if exist p and q to satisfy p＜q, de
P-1＞=0, de
Q+1＞=0, and de
p=de
P+1=... .=de
Q-1=-1, then the border between p and the q is a positive straight line, and its length is q-p.
Rule 3: if exist p and q to satisfy p＜q, de
P-1＜=0, de
Q+1＜=0,, and de
p=de
P+1=... .=de
Q-1=-1, then the border between p and the q is a negative straight line, and its length is q-p.
Step 5 is represented border E with following S set,
S＝{(t
m，s
m，e
m，l
m)|m＝1，2，....，M}，
Wherein, t
mBe the type of m ripple unit among the E, value is 0,1,2, respectively expression zero straight line, positive straight line and negative straight line; s
m, e
mAnd l
mBe respectively starting point, terminal point and the length of m ripple unit; M is the number of E medium wave unit.
Step 6 merges continuous wave unit according to rule 4,5,6.
Rule 4: if t
p=t
P+1, | E (e
p)-E (s
P+1) |＜3, and s
P+1-e
p＜2, then these two ripple units are merged, with seasonal M=M-1.
Rule 5: if t
p=t
P+2, | E (e
p)-E (s
P+2) |＜3, and l
P+1＜3, then p, p+1, three ripple units of p+2 are merged, with seasonal M=M-2.
Rule 6: if
Then with this ripple unit deletion, with seasonal M=M-1.
Then, from through extracting following recognition feature the above-mentioned processing resulting alphabetical border set: the number of straight line just in the length of the longest zero straight line, the coboundary in the length of the longest zero straight line, the lower boundary in the length of the longest zero straight line, the right margin in the length of article one zero straight line, the coboundary in the length of article one zero straight line, the lower boundary in the zero straight line number in the depth-width ratio of alphabetical depth-width ratio (HW), alphabetical slave part, the unit of the ripple in each border number, each border, the right margin.
Present embodiment adopts Fig. 1～decision tree shown in Figure 4 to discern the print hand Arabic alphabets of absolute version, beginning form, intermediate form, ending form respectively.During identification, at first, utilize alphabetical depth-width ratio that Arabic alphabet is divided into three classes, then for alphabetical number greater than 1 classification, utilize the feature of being extracted that it is further segmented, till identifying this letter.With Fig. 1 is example, and the Arabic alphabet recognition rule of absolute version is as follows:
R1. if HW＜1, and LN4 ≠ 1, then be letter b aa;
R2. if HW＜1, and LN4=1, and 0.5W＜MSL4＜0.7W, and LN3=2, then be alphabetical Sad;
R3. if HW＜1, and LN4=1, and 0.5W＜MSL4＜0.5W, and LN3 ≠ 2, then be alphabetical Dhad;
R4. if HW＜1, and LN4=1, and MSL4＜=0.5W, and LN3 ≠ 2, then be alphabetical Sheen;
R5. if HW＜1, and LN4=1, and MSL4＜=0.5W, and LN3=2, then be alphabetical Seen;
R6. if HW＜1, and LN4=1, and MSL4＞=0.7W, and MSL2＞0.4W, then be alphabetical Faa;
R7. if HW＜1, and LN4=1, and MSL4＞=0.7W, and MSL2＜=0.4W, and SHW＞0.9, then be tee haa;
R8. if HW＜1, and LN4=1, and MSL4＞=0.7W, and MSL2＜=0.4W, and SHW＜=0.9, then be tee aa;
R9. if HW＞3 then are alphabetical Alif;
R10. if 1＜HW＜3, and LN1=LN3=LN4=1, and MSL4＞0.7W, then be alphabetical Dal;
R11. if 1＜HW＜3, and LN1=LN3=LN4=1, and MSL4＜=0.7W, and LN2 ≠ 1, then be letter w aw;
R12. if 1＜HW＜3, and LN1=LN3=LN4=1, and MSL4＜=0.7W, and LN2=1, and MSL2＞0.7W, then be alphabetical Hha;
R13. if 1＜HW＜3, and LN1=LN3=LN4=1, and MSL4＜=0.7W, and LN2=1, and MSL2＜=0.7W, and SN2=1, then be letter r aa;
R14. if 1＜HW＜3, and LN1=LN3=LN4=1, and MSL4＜=0.7W, and LN2=1, and MSL2＜=0.7W, and SN2 ≠ 1, then be alphabetical Ha;
R15. if 1＜HW＜3, and LN3=LN4=1, and LN1 ≠ 1, and MSL2＞0.4W, then be alphabetical Lam;
R16. if 1＜HW＜3, and LN3=LN4=1, and LN1 ≠ 1, and MSL2＜=0.4W, then be alphabetical Kaf;
R17. if 1＜HW＜3, and LN1=LN4=1, and LN3 ≠ 1, and MSL2＞0.8W, then be alphabetical Jeem;
R18. if 1＜HW＜3, and LN1=LN4=1, and LN3 ≠ 1, and MSL2＜=0.8W, and MSL4＞0.8W, and MSL3＞0.5H then is tee ta;
R19. if 1＜HW＜3, and LN1=LN4=1, and LN3 ≠ 1, and MSL2＜=0.8W, and MSL4＞0.8W, and MSL3＜=0.5H, then be tee tha;
R20. if 1＜HW＜3, and LN1=LN4=1, and LN3 ≠ 1, and MSL2＜=0.8W, and MSL4＜=0.8W, then be alphabetical Ain;
R21. if 1＜HW＜3, and LN4=1, and LN1 ≠ 1, and LN3 ≠ 1, and MSL4＞0.8W, then be tee hal;
R22. if 1＜HW＜3, and LN4=1, and LN1 ≠ 1, and LN3=2, and 0.4W＜MSL4＜=0.8W, and MSL3＞0.6H, and PN2=1, then be alphabetical Gaf;
R23. if 1＜HW＜3, and LN4=1, and LN1 ≠ 1, and LN3=2, and 0.4W＜MSL4＜=0.8W, and MSL3＞0.6H, and PN2 ≠ 1, then be alphabetical Noon;
R24. if 1＜HW＜3, and LN4=1, and LN1 ≠ 1, and LN3=2, and 0.4W＜MSL4＜=0.8W, and MSL3＜=0.6H, then be alphabetical Meem;
R25. if 1＜HW＜3, and LN4=1, and LN1 ≠ 1, and LN3=2, and MSL4＜=0.4W then is zed ay;
R26. if 1＜HW＜3, and LN4=1, and LN1 ≠ 1, and LN3 ≠ 2, and LN3 ≠ 1, and SN1=2, then be alphabetical Ghain;
R27. if 1＜HW＜3, and LN4=1, and LN1 ≠ 1, and LN3 ≠ 2, and LN3 ≠ 1, and SN1 ≠ 2, then be alphabetical Kha;
R28. if 1＜HW＜3, and LN4 ≠ 1, and LN3 ≠ 1, then be alphabetical Yaa;
R29. if 1＜HW＜3, and LN4 ≠ 1, and LN3=1, then be alphabetical Lam-Alif.
In conjunction with Fig. 2-Fig. 4, the recognition methods of the Arabic alphabet of beginning, middle and ending form is identical with the recognition methods of absolute version.
Claims (7)
1. the method for recognizing print hand Arabic alphabets based on boundary characteristic is characterized in that: the first step, four borders, upper and lower, left and right of calculating letter; Second step was the set of a series of ripple unit with each boundary representation; In the 3rd step, from the set of ripple unit, extract ripple unit number, zero straight line number, the longest zero straight length as recognition feature; At last, adopt decision tree to discern each print hand Arabic alphabets.
2. the method for recognizing print hand Arabic alphabets based on boundary characteristic according to claim 1 is characterized in that: described four borders, upper and lower, left and right of calculating letter be adopt from top to bottom, from top to bottom, from left to right, the mode of four direction scanning from right to left four borders of calculating letter.
3. the method for recognizing print hand Arabic alphabets based on boundary characteristic according to claim 2 is characterized in that: the computing method on alphabetical border are as follows: left margin from left to right scans letter image, establishes e
pThe number of the continuous white pixel when running into first black pixel for scanning from letter image p capable left side beginning to the right, then Zi Mu left margin E can be expressed as: E={e
p| p=1,2 ...., N}, wherein N is a letter height, and the computing method on its excess-three border are identical, are the direction of scanning difference, and the direction of scanning of coboundary is for from top to bottom, and right margin is for from right to left, and lower boundary is for from top to bottom.
4. the method for recognizing print hand Arabic alphabets based on boundary characteristic according to claim 1, it is characterized in that: with four boundary representations up and down of letter is the set of zero straight line, positive straight line, negative three kinds of ripple units of straight line, and to its merge, deletion action.
5. the method for recognizing print hand Arabic alphabets based on boundary characteristic according to claim 4 is characterized in that: with each boundary representation is that the step that ripple unit gathers is: the step that left margin is expressed as the set of ripple unit is:
Step 1, the difference dE={de of computation bound E
p| p=2,3 ..., N}, wherein de
pComputing method be
Step 2, dE is carried out smoothly:
De
p=de
P-1If de
P-1=de
P+1
Step 3, utilize rule 1,2,3 to find in the border three types ripple unit---zero straight line, positive straight line and negative straight line:
Rule 1: if exist p and q to satisfy p＜q, de
P-1≠ 0, de
Q+1≠ 0, and de
p=de
P+1=... .=de
Q-1=0, then the border between p and the q is one zero straight line, and its length is q-p;
Rule 2: if exist p and q to satisfy p＜q, de
P-1＞=0, de
Q+1＞=0, and de
p=de
P+1=... .=de
Q-1=-1, then the border between p and the q is a positive straight line, and its length is q-p;
Rule 3: if exist p and q to satisfy p＜q, de
P-1＜=0, de
Q+1＜=0,, and de
p=de
P+1=... .=de
Q-1=-1, then the border between p and the q is a negative straight line, and its length is q-p;
Step 4 is if the length of ripple unit then writes down type, starting point, end point and the length thereof of this ripple unit greater than 2;
Step 5 is represented border E with following S set,
S＝{t
m，s
m，e
m，l
m)|m＝1，2，...，M}，
Wherein, t
mBe the type of m ripple unit among the E, value is 0,1,2, respectively expression zero straight line, positive straight line and negative straight line; s
m, e
mAnd l
mBe respectively starting point, terminal point and the length of m ripple unit; M is the number of E medium wave unit;
Step 6 merges continuous wave unit according to rule 4,5,6:
Rule 4: if t
p=t
P+1, | E (e
p)-E (s
P+1) |＜3, and s
P+1-e
p＜2, then these two ripple units are merged, with seasonal M=M-1;
Rule 5: if t
p=t
P+2, | E (e
p)-E (s
P+2) |＜3, and l
P+1＜3, then p, p+1, three ripple units of p+2 are merged, with seasonal M=M-2;
Rule 6: if
Then with this ripple unit deletion, with seasonal M=M-1;
The computing method on its excess-three border are identical.
6. the method for recognizing print hand Arabic alphabets based on boundary characteristic according to claim 1, it is characterized in that: it is at first to utilize alphabetical depth-width ratio that print hand Arabic alphabets is divided three classes that described employing decision tree discerns each print hand Arabic alphabets, for each class, utilize following feature again: the depth-width ratio of alphabetical slave part, ripple unit number in each border, zero straight line number in each border, the length of article one zero straight line in the right margin, the length of article one zero straight line in the lower boundary, the length of the longest zero straight line in the coboundary, the length of the longest zero straight line in the right margin, the number of positive straight line in the length of the longest zero straight line and the coboundary in the lower boundary, and in conjunction with alphabetical depth-width ratio, the depth-width ratio of letter slave part is jointly as recognition feature identification print hand Arabic alphabets.
7. the method for recognizing print hand Arabic alphabets based on boundary characteristic according to claim 6 is characterized in that: according to four kinds of forms of letter: independent, beginning, middle, ending, and adopt four decision trees to discern each print hand Arabic alphabets respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007100721630A CN100492404C (en) | 2007-04-30 | 2007-04-30 | Method for recognizing print hand Arabic alphabets based on boundary characteristic |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2007100721630A CN100492404C (en) | 2007-04-30 | 2007-04-30 | Method for recognizing print hand Arabic alphabets based on boundary characteristic |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101038627A true CN101038627A (en) | 2007-09-19 |
| CN100492404C CN100492404C (en) | 2009-05-27 |
Family
ID=38889516
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2007100721630A Expired - Fee Related CN100492404C (en) | 2007-04-30 | 2007-04-30 | Method for recognizing print hand Arabic alphabets based on boundary characteristic |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100492404C (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102142088A (en) * | 2010-08-17 | 2011-08-03 | 穆罕默德S·卡尔希德 | Effective Arabic feature extraction-based Arabic identification method and system |
-
2007
- 2007-04-30 CN CNB2007100721630A patent/CN100492404C/en not_active Expired - Fee Related
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102142088A (en) * | 2010-08-17 | 2011-08-03 | 穆罕默德S·卡尔希德 | Effective Arabic feature extraction-based Arabic identification method and system |
| CN102142088B (en) * | 2010-08-17 | 2013-01-23 | 穆罕默德S·卡尔希德 | Effective Arabic feature extraction-based Arabic identification method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN100492404C (en) | 2009-05-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110569832B (en) | Text real-time positioning and identifying method based on deep learning attention mechanism | |
| CN1284071C (en) | Recognition trigger with pen when near and hand wiriting when far | |
| CN106650664A (en) | Collection system and method for college and university enrollment book data | |
| CN101038627A (en) | Method for recognizing print hand Arabic alphabets based on boundary characteristic | |
| CN1263302A (en) | Pen and signal based manuscript editing technique | |
| CN103902100A (en) | Speed characteristics based stroke partitioning method for intelligent freehand sketching | |
| CN1530804A (en) | Rapid Chinese handwritnig inputting method | |
| CN1734467A (en) | Mobile phone capable of inputting text by taking pictures and inputting method thereof | |
| KR20090111202A (en) | The Optical Character Recognition method and device by the numbers of horizon, vertical and slant lines which is the element of Hanguel | |
| CN1262488A (en) | Braille generation method and its reader/typewriter | |
| CN110852359A (en) | Family tree identification method and system based on deep learning | |
| CN87100210A (en) | Coding method of Chinese characters using first and final strokes | |
| CN1191548C (en) | Extracting method for Chinese character stroke based on sectional law | |
| CN113506278B (en) | Road sign auxiliary detection control method based on YOLO-V3 | |
| CN1088877C (en) | Photographic intelligent reading device and its operating method | |
| CN85100084A (en) | Real-time distinguishing apparatus for microcomputer hand written chinese character | |
| CN1731326A (en) | Double-stroke input method of Chinese character for computer and handset keyboard | |
| CN1133134C (en) | Method and device for automatically extracting strokes of Chinese characters | |
| CN1048561C (en) | Method for inputting Chinese words into computer | |
| CN1200375C (en) | Content filter training and estimating method | |
| CN1162158A (en) | Method for automatically correcting truncating error of document and device thereof | |
| CN1033188C (en) | Chinese-character stroke encoding method | |
| CN1006419B (en) | Sorting and coding method of chinese character outline symbol and character element | |
| JP5986051B2 (en) | Method for automatically recognizing Arabic text | |
| CN1073723C (en) | Chinese character keyboard input method based on configuration |
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 | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090527 Termination date: 20120430 |