CN101763516A - Character recognition method based on fitting functions - Google Patents
Character recognition method based on fitting functions Download PDFInfo
- Publication number
- CN101763516A CN101763516A CN201010017933A CN201010017933A CN101763516A CN 101763516 A CN101763516 A CN 101763516A CN 201010017933 A CN201010017933 A CN 201010017933A CN 201010017933 A CN201010017933 A CN 201010017933A CN 101763516 A CN101763516 A CN 101763516A
- Authority
- CN
- China
- Prior art keywords
- point
- stroke
- character
- identified
- line segment
- 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
- 238000000034 method Methods 0.000 title claims abstract description 56
- 230000006870 function Effects 0.000 title abstract description 40
- 239000013598 vector Substances 0.000 claims description 47
- 230000015572 biosynthetic process Effects 0.000 claims description 7
- 206010044565 Tremor Diseases 0.000 claims description 6
- 230000004907 flux Effects 0.000 claims description 5
- 238000005194 fractionation Methods 0.000 claims description 5
- 230000008878 coupling Effects 0.000 claims description 3
- 238000010168 coupling process Methods 0.000 claims description 3
- 238000005859 coupling reaction Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000009792 diffusion process Methods 0.000 claims description 2
- 238000010187 selection method Methods 0.000 claims description 2
- 238000009826 distribution Methods 0.000 description 5
- 230000000875 corresponding effect Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000000052 comparative effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 230000002596 correlated effect Effects 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 230000013011 mating Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
Images
Landscapes
- Character Discrimination (AREA)
Abstract
The invention discloses a character recognition method based on fitting functions, which comprises the following steps of: fitting lines forming a character to be recognized, converting the character into a plurality of polynomial functions and describing and then taking out a template with the highest similarity as a recognition result by analyzing the similarity of fitting functions and standard functions of a template base. The invention not only can recognize Chinese characters but also can recognize numbers, letters and some special symbols and has higher executing efficiency and accuracy.
Description
Technical field
The invention belongs to pattern-recognition, artificial intelligence and Word message process field, particularly a kind of character recognition method based on fitting function.
Background technology
Along with applying of computer technology, especially the internet is universal day by day, and the mankind rely on computing machine to obtain various information more and more.Information processing work is also all transferred on the computing machine and is carried out in large quantities.In daily live and work, exist a large amount of Word message and handle problems, thereby Word message is input to this requirement of computing machine fast becomes very urgent.
General literal recognizer is based on the dot-matrix analysis to figure to be identified now, significant limitation discerned in the complicated literal of lines that constitutes literal, for example Chinese character there is special Chinese characters recognition method, letter is had special alphabetical recognition methods, and promptly using same class algorithm thought also is to need different programs to realize.When especially the polytype symbol being mixed, most of algorithm all is difficult to identify.If handwriting input, the difficulty of identification is just bigger.
Existing common language recognizer, for example template matching algorithm is generally described with the mean value of training sample feature, and sorter is discerned according to the distance of the reference feature of input sample characteristics and each literal.Because there is various font in Chinese character, also have various distortion in the handwritten Chinese character, so all there is a distribution space in any feature of file.Have only these distributions are taken into account, could carry out Classification and Identification more accurately, it is not enough therefore only describing feature with feature mean value.
The Bayes algorithm is to adopt a kind of probability density function method of describing Hanzi features with piecewise linear function, can well identify the characteristic distribution of Chinese character, but because the distribution of eigenwert is not certain simple statistical distribution usually, if there is not simple method to describe these probability density functions, the storage space of probability density function needs of then representing each dimensional feature of all Chinese characters will be utility system can not bear.
More than these two kinds of methods all can not be well with the complex figure classification, identification is carried out to literal accuracy rate and efficient are very low.
Summary of the invention
The present invention seeks to defective,, use the technology and the feature of fitting function, propose a kind of effective character recognition method based on fitting function according to the characteristics of character image at the prior art existence.The present invention has better taken into account the accuracy and the speed of literal identification.
The present invention adopts following technical scheme for achieving the above object:
The present invention is a kind of character recognition method based on fitting function, it is characterized in that comprising the steps:
(1) under online or situation, reads in literal sample to be identified at off-line;
(2) character image pre-service: at first with the described literal sample image binaryzation to be identified of step (1), lines disappear and tremble then, carry out breakpoint again and connect, and at last character image are standardized;
(3) analyzing samples: split out strokes of characters earlier from the literal sample image to be identified after the described standardization of step (2), each stroke that will split out then is referred in the stroke set, and generates discernible symbol sebolic addressing with this; Every in stroke set line segment is sought unique point, find out the best-fit function according to the feature point set polynomial fitting, the coefficient of choosing the best-fit function is as the best-fit vector; Write symbolic vector with best-fit to flux matched stroke;
(4) recognition sample: at first in knowledge base, seek a plurality of templates that match according to the described symbol sebolic addressing of step (3); Extract the symbolic vector of these a plurality of die plate pattern then and mate with the symbolic vector of the described literal sample image to be identified of step (3);
(5) output result: that will be with the symbolic vector similarity of the described literal sample image to be identified of step (3) the highest template character pattern is as recognition result.
The present invention be directed to the identification of two-value character image, the image after the binaryzation can make image processing velocity height, cost low, in addition, owing to be that character string is mated finally, makes index of reference improve matching speed, therefore has higher execution efficient; The present invention is converted into polynomial function with the character and graphic of various complexity and is described, and generates symbolic vector, then by analyze with knowledge base in the similarity of template carry out literal identification, to literal recognition correct rate height, good stability.
Description of drawings
Fig. 1: the The general frame of the inventive method;
Fig. 2: stroke splits process flow diagram under the online situation.
Embodiment
For the ease of narrating the character recognition method based on fitting function provided by the invention, at first provide about the following basic definition of this method:
Define a line segment (LS, Line Segment) and line-segment sets (LSS, Line Segment Set):
One section continuous, is made of some spots, can be called line segment (can be segment of curve or straight-line segment) by the point sequence of match, and the sequence of being made up of line segment is called line-segment sets.
Definition: LS=((X
1, Y
1), (X
2, Y
2) ..., (X
Npi, Y
Npi) ..., (X
Np, Y
Np)), Np is the quantity of point.
LSS=(LS
1, LS
2..LS
Nli..., LS
Nl), Nl is the quantity of line segment.
Wherein: Np
i, Nl
i, Np, Nl ∈ Z+, X
Npi, Y
Npi∈ Z, LS ∈ LSS, | X
Npi-X
Npi-1|≤1, | Y
Npi-Y
Npi-1|≤1.
Define two unique points (LC, Line Character points) and feature point set (LCS, Line Character pointsSet):
Chosen wherein Ncp and put and carry out match by Np the line segment LS that constitute of point for one, each point that is selected is called the unique point of line segment LS, and the sequence of being made up of this Ncp unique point is called feature point set.
Definition: LC ∈ LS.
LCS={LC
1, LC
2..., LC
Ncpi..., LC
Ncp, Ncp is the quantity of unique point.
Wherein: LC
Ncpi∈ LS, 0≤Ncp
i≤ Ncp≤Np.
Define three best-fit functions (BFF, Best-Fitting Function):
A feature point set LCS who utilizes a line segment LS carries out the fitting function that p rank polynomial expression Y (X) that match obtains or X (Y) are called this line segment to line segment, and wherein Zui You fitting function is called the best-fit function.
Definition: BFF (X)=A
p* X
p+ A
P-1* X
P-1+ ... A
j* X
j+ ...+A
0* X
0, Flag zone bit 0.
BFF (Y)=A
p* Y
n+ A
P-1* Y
P-1+ ... A
j* X
j+ ...+A
0* Y
0, Flag zone bit 1.
Wherein: A
j∈ R, j, p ∈ Z
+
Define four best-fit vectors (BFV, Best-Fitting Vector):
The coefficient A of best-fit function
jAnd the vector that the zone bit Flag of Y (X) type or X (Y) type forms is called the best-fit vector.
Definition: BFV=(Flag, A
p, A
P-1..., A
j... A
0).
Wherein: A
j∈ R, j, p ∈ Z
+
Define five strokes (SK, Strokes):
Be called stroke by some vectorial set of forming of best-fit, but be that the element institute that distinguished symbol is concentrated can divided least unit with like attribute.
Definition: SK={BFV
1, BFV
2..., BFV
Nski..., BFV
Nsk, Nsk is the template samples number of stroke.
Wherein: Nsk
i, Nsk ∈ Z
+
For example: the sets definitions with the approximately parallel match vector of X-axis formation all in the Chinese character are " horizontal stroke " this stroke.
Definition six-stroke collection (SKS, Strokes Set):
The set by but distinguished symbol concentrates all strokes to be constituted is called stroke set.
Definition: SKS={SK
1, SK
2..., SK
Nsksi..., SK
Nsks, Nsks is the quantity of stroke.
Wherein: Nsks
i, Nsks ∈ Z
+
For example: Chinese character is made of the set that " horizontal stroke " " erects " " left-falling stroke " a series of basic strokes such as " right-falling strokes ".
Define seven symbolic vectors (SV, Symbol Vector):
But be called symbolic vector by the vector to be identified or distinguished symbol of can representing that n stroke element constitutes.The image information that can comprise a pictorial element record symbol in case of necessity is so that identification more accurately.
Definition: SV=(SK
1, SK
2..., Sk
Nsvi..., SK
Nsv), Nsv is the stroke number that constitutes this symbol.
Wherein: Nsv
i, Nsv ∈ Z
+
For example: Chinese character " ten " be made up of " horizontal stroke " stroke and " an erecting " stroke that a Chinese-character stroke is concentrated, thereby " ten " this symbol is made of (horizontal stroke, perpendicular) this vector.
Define eight glossary of symbols (SVS, Symbol Vector Set):
But the set that is made of all distinguished symbol vectors is called glossary of symbols.
Definition: SVS={SV
1, SV
2..., SV
Nsvsi..., SV
Nsvs, but Nsvs is the quantity of distinguished symbol vector.
Wherein: Nsvs
i, Nsvs ∈ Z
+
Define nine knowledge bases (KB, Knowledge Base):
The template base of being made up of stroke set template and glossary of symbols template is called knowledge base.
Definition: KB=＜SKS, SVS 〉.
Define ten similarities (Smlt, Similarity):
The similarity degree of glossary of symbols template is called similarity in sample to be identified and the knowledge base.
Provided by the inventionly be based on the concrete performing step of the character recognition method of fitting function: input literal sample to be identified under online or situation at off-line, then to the character image sample of input carry out binaryzation, lines disappear tremble, pre-service work that breakpoint connections, standardization etc. are correlated with; Then carry out fractionation, the selected characteristic point of line segment, utilize the feature point set LCS of a line segment LS that line segment is carried out match, obtain best fitting function and best-fit vector, and then from figure, split out the pattern of stroke, the stroke pattern that splits out analyzed be referred in the stroke set the stroke of coupling, and generate discernible symbol sebolic addressing with this; Then in knowledge base, find out several templates of mating most with it, and extract the pattern of these several templates and sample to be identified mates, calculate the similarity degree of these several templates in sample to be identified and the knowledge base at last respectively, literal that wherein similarity is the highest final recognition result as literal to be identified.
Detailed process is as follows:
As shown in Figure 1.A kind of character recognition method based on fitting function is characterized in that comprising the steps:
(1) under online or situation, reads in literal sample to be identified at off-line;
(2) character image pre-service: at first with the described literal sample image binaryzation to be identified of step (1), lines disappear and tremble then, carry out breakpoint again and connect, and at last character image are standardized;
(3) analyzing samples: split out strokes of characters earlier from the literal sample image to be identified after the described standardization of step (2), each stroke that will split out then is referred in the stroke set, and generates discernible symbol sebolic addressing with this; Every in stroke set line segment is sought unique point, find out the best-fit function according to the feature point set polynomial fitting, the coefficient of choosing the best-fit function is as the best-fit vector; Write symbolic vector with best-fit to flux matched stroke;
(4) recognition sample: at first in knowledge base, seek a plurality of templates that match according to the described symbol sebolic addressing of step (3); Extract the symbolic vector of these a plurality of die plate pattern then and mate with the symbolic vector of the described literal sample image to be identified of step (3);
(5) output result: that will be with the symbolic vector similarity of the described literal sample image to be identified of step (3) the highest template character pattern is as recognition result.
Described a kind of character recognition method based on fitting function, it is characterized in that step (5) output result after, the described literal sample image to be identified of step (3) is saved as template.
Described a kind of character recognition method based on fitting function is characterized in that the described binarization method of step (2) is as follows:
At first get T
1=70 as the threshold value of selecting for use for the first time, and the described literal sample image to be identified of step (1) is carried out the global threshold binaryzation, distinguishes the background and the image of literal sample image to be identified; Obtain corresponding threshold value T more respectively
1The image of literal sample image to be identified and the average gray F of background
aAnd B
a, make the threshold value that secondary selects for use and be:
The last threshold value T that selects for use with secondary again
2Former figure is carried out the global threshold binaryzation.
Described a kind of character recognition method based on fitting function is characterized in that the described lines of step (2) fluttering method that disappears is as follows:
Randomly draw the point that accounts for total strong point 50% in the text handwriting in the literal sample image to be identified after binaryzation, connect the consecutive point that nearest point and distance are no more than 3 pixels with straight line.
Described a kind of character recognition method based on fitting function is characterized in that the method for attachment of the described literal breakpoint of step (2) is as follows:
Randomly draw the point that accounts for total strong point 60% in the text handwriting in lines disappear literal sample image to be identified after trembling, connect the consecutive point that nearest point and distance are no more than 3 pixels with straight line.
As shown in Figure 2.Described a kind of character recognition method based on fitting function is characterized in that the method for the described fractionation strokes of characters of step (3) is as follows:
Online situation:
1. two of initialization cover table CovX and CovY, be respectively applied for the situation that is capped of record X-axis and Y-axis, and all are changed to False, represent that two covering table CovX and CovY are not capped; Array LS of initialization is used for storing the line segment that fractionation is come out, Nl
j=0;
2. set up two empty queue QueueX and QueueY, be respectively applied for and be pressed into the point sequence that writes down and write down by X (Y) mode by Y (X) mode;
3. whenever obtaining a point (X
i, Y
i) time, judge X respectively at two covering table CovX and CovY
iPosition and Y
iWhether the position is False: if the X of CovX
iThe position is False, then with (X
i, Y
i) be pressed into formation QueueX and at the X of covering table CovX
iThe position writes True; If cover the Y of table CovY
iThe position is that False is then with (X
i, Y
i) be pressed into formation QueueY and at the Y of covering table CovY
iThe position writes True; A Cov table in running into two covering table CovX and CovY table is prominent to be that correspondence position is the X of True
iOr Y
iThe position has been True, then enters step 4.; All handle when point, then enter step 5.;
If 4. the conflict of Cov table is arranged, then the whole dequeues of the element among the corresponding queues Queue are emptied and the Cov table is resetted; Another Cov table is proceeded the operation of step in 3., up to conflict also occurring, then with the whole dequeues of corresponding formation Queue to line segment LS[Nl
j], Nl
j=Nl
j+ 1, empty the Cov table;
5. if a strokes of characters line drawing is intact, then enters step and 1. split next bar strokes of characters; If point can not obtain, then finish;
Off-line case:
A. according to from left to right, scan method from top to bottom as long as scan a point, just with the method diffusion of this point with se ed filling algorithm, all is communicated with the line segment that can be communicated with, and the method with online identification generates LS then, changes step (B);
B. whenever take a line segment away, judge whether this line segment intersects with other line segments,, then fill the point of crossing, change step (A), till in figure, not put if any intersecting if do not intersect then from figure, delete this line segment.
Its mid point (X
i, Y
i) expression constitutes the i point of a strokes of characters, i ∈ Z
+, 0≤i≤Np, Np are counting of a strokes of characters.
Described a kind of character recognition method based on fitting function is characterized in that the described feature point selection method of step (3) is as follows:
Starting point (the X of line segment
1, Y
1) and terminal point (X
Np, Y
Np) should be selected, remaining point should be got from the various piece of line segment with certain rule, and the present invention gets a unique point every 5 points, and the assurance unique point can be described the profile of former line segment preferably.
Described a kind of character recognition method based on fitting function is characterized in that the method for the described definite best-fit vector of step (3) is as follows:
Determine exponent number p and coefficient A in the best-fit function
j, j=1,2...p, the coefficient of choosing the best-fit function then is the best-fit vector the most;
Exponent number p obtains in the following manner in the best-fit function: p obtains match similarity R since 1 match
2, whenever p+1 can make match similarity R
2Increase threshold alpha=5%, then getting p+1 is current p, judges once more, can not make match similarity R up to the match of p+1 rank
2Increase till the α;
Each coefficient A in the best-fit function
jObtain in the following manner: for coefficient A
jIf, A
jAbsolute value less than a threshold value beta=1E-3, A is removed in test so
j* X
jAfter, match similarity R
2Can or can not reduce threshold gamma=5%,, just remove A if it is so much can not to reduce γ
j* X
jItem is promptly A
jAssignment is 0.
Described a kind of character recognition method based on fitting function is characterized in that the highest template of the described searching similarity of step (4) is as follows as the method for recognition result:
(a) each best-fit vector that splits out according to figure is searched the stroke that mates most in the stroke set of knowledge base, generate symbolic vector;
(b) utilize symbolic vector in the glossary of symbols of knowledge base, to search the symbol of coupling, if stroke be orderly so in search procedure stroke and order all to mate; Search according to the quantity and the type of stroke so if stroke is unordered, do not consider to mate in proper order;
(c) if under orderly with the unordered situation of stroke, can not find stroke number identical can not find used stroke identical, then use near matching principle, look for immediate symbol:, then get each and can match the maximum character of correct stroke quantity in order for the orderly character vector of stroke; If stroke is unordered, then get and can match the maximum character of used stroke in all strokes, then as required, can take out the image of character vector, compare the affirmation result with image.
The implementation of algorithm of the present invention is:
S1.KB=CreatKB () or LoadKB (KB); // set up knowledge base or be written into knowledge base
S2.SV.Pic=ReadSV (); // read in sample
S3.AnalysisSV (SV) // analyzing samples
Begin
LSS=GetLSS (SV.Pic); // image is split into stroke set
For?k：＝1?To?LSS.size()Do
Begin // generating feature point set
LCS=GetLCS (LSS[k]); // generate the best-fit vector to write symbolic vector
SV.SKV[k] .BFV.add (GetBFV (LCS)); // use best-fit to flux matched stroke
SV.SKV[k]＝MatchingSK(SV.SKV[k].BFV)；
End
return?SUCCESS；
End
S4.SVS=MatchingTP (SV); // matching template
S5.Begin // press to flux matched several symbolic vectors that get
SVS＝MatchingSV(SV.SKV)；
// if desired, the image of the image of comparative sample and each possible symbol mates
SVS＝MatchingPIC(SV.Pic)；
If (user need refresh one's knowledge storehouse)
Then
StoreKB(SV)；
return?SVS；
End
Write (SVS[0]); The result that // output is mated most
Claims (9)
1. the character recognition method based on fitting function is characterized in that comprising the steps:
(1) under online or situation, reads in literal sample to be identified at off-line;
(2) character image pre-service: at first with the described literal sample image binaryzation to be identified of step (1), lines disappear and tremble then, carry out breakpoint again and connect, and at last character image are standardized;
(3) analyzing samples: split out strokes of characters earlier from the literal sample image to be identified after the described standardization of step (2), each stroke that will split out then is referred in the stroke set, and generates discernible symbol sebolic addressing with this; Every in stroke set line segment is sought unique point, find out the best-fit function according to the feature point set polynomial fitting, the coefficient of choosing the best-fit function is as the best-fit vector; Write symbolic vector with best-fit to flux matched stroke;
(4) recognition sample: at first in knowledge base, seek a plurality of templates that match according to the described symbol sebolic addressing of step (3); Extract the symbolic vector of these a plurality of die plate pattern then and mate with the symbolic vector of the described literal sample image to be identified of step (3);
(5) output result: that will be with the symbolic vector similarity of the described literal sample image to be identified of step (3) the highest template character pattern is as recognition result.
2. a kind of character recognition method according to claim 1 based on fitting function, it is characterized in that step (5) output result after, the described literal sample image to be identified of step (3) is saved as template.
3. a kind of character recognition method based on fitting function according to claim 1 is characterized in that the described binarization method of step (2) is as follows:
At first get T
1=70 as the threshold value of selecting for use for the first time, and the described literal sample image to be identified of step (1) is carried out the global threshold binaryzation, distinguishes the background and the image of literal sample image to be identified; Obtain corresponding threshold value T more respectively
1The image of literal sample image to be identified and the average gray F of background
aAnd B
a, make the threshold value that secondary selects for use and be:
The last threshold value T that selects for use with secondary again
2Former figure is carried out the global threshold binaryzation.
4. a kind of character recognition method based on fitting function according to claim 1 is characterized in that the described lines of step (2) fluttering method that disappears is as follows:
Randomly draw the point that accounts for total strong point 50% in the text handwriting in the literal sample image to be identified after binaryzation, connect the consecutive point that nearest point and distance are no more than 3 pixels with straight line.
5. a kind of character recognition method based on fitting function according to claim 1 is characterized in that the method for attachment of the described literal breakpoint of step (2) is as follows:
Randomly draw the point that accounts for total strong point 60% in the text handwriting in lines disappear literal sample image to be identified after trembling, connect the consecutive point that nearest point and distance are no more than 3 pixels with straight line.
6. a kind of character recognition method based on fitting function according to claim 1 is characterized in that the method for the described fractionation strokes of characters of step (3) is as follows:
Online situation:
1. two of initialization cover table CovX and CovY, be respectively applied for the situation that is capped of record X-axis and Y-axis, and all are changed to False, represent that two covering table CovX and CovY are not capped; Array LS[Nl of initialization
j], be used for storing the line segment that fractionation is come out, Nl
j=0;
2. set up two empty queue QueueX and QueueY, be respectively applied for and be pressed into the point sequence that writes down and write down by X (Y) mode by Y (X) mode;
3. whenever obtaining a point (X
i, Y
i) time, judge X respectively at two covering table CovX and CovY
iPosition and Y
iWhether the position is False: if the X of CovX
iThe position is False, then with (X
i, Y
i) be pressed into formation QueueX and at the X of covering table CovX
iThe position writes True; If cover the Y of table CovY
iThe position is that False is then with (X
i, Y
i) be pressed into formation QueueY and at the Y of covering table CovY
iThe position writes True; A Cov table in running into two covering table CovX and CovY table is prominent to be that correspondence position is the X of True
iOr Y
iThe position has been True, then enters step 4.; All handle when point, then enter step 5.;
If 4. the conflict of Cov table is arranged, then the whole dequeues of the element among the corresponding queues Queue are emptied and the Cov table is resetted; Another Cov table is proceeded the operation of step in 3., up to conflict also occurring, then with the whole dequeues of corresponding formation Queue to line segment LS[Nl
j], Nl
j← Nl
j+ 1, empty the Cov table, wherein ← the expression assignment;
5. if a strokes of characters line drawing is intact, then enters step and 1. split next bar strokes of characters; If point can not obtain, then finish;
Off-line case:
A. according to from left to right, scan method from top to bottom as long as scan a point, just with the method diffusion of this point with se ed filling algorithm, all is communicated with the line segment that can be communicated with, and the method with online identification generates LS then, changes step B;
B. whenever take a line segment away, judge whether this line segment intersects with other line segments,, then fill the point of crossing, change steps A, till in figure, not put if any intersecting if do not intersect then from figure, delete this line segment;
Its mid point (X
i, Y
i) expression constitutes the i point of a strokes of characters, i ∈ Z
+Be positive integer, 0≤i≤Np, Np are counting of a strokes of characters.
7. a kind of character recognition method based on fitting function according to claim 1 is characterized in that the described feature point selection method of step (3) is as follows:
Starting point (the X of line segment
1, Y
1) and terminal point (X
Np, Y
Np) should be selected, remaining point should be got from the various piece of line segment with certain rule, and the present invention gets a unique point every 5 points, and the assurance unique point can be described the profile of former line segment preferably.
8. a kind of character recognition method based on fitting function according to claim 1 is characterized in that the method for the described definite best-fit vector of step (3) is as follows:
Determine exponent number p and coefficient A in the best-fit function
j, j=1,2...p, the coefficient of choosing the best-fit function then is the best-fit vector the most;
Exponent number p obtains in the following manner in the best-fit function: p obtains match similarity R since 1 match
2, whenever p+1 can make match similarity R
2Increase threshold alpha=5%, then getting p+1 is current p, judges once more, can not make match similarity R up to the match of p+1 rank
2Increase till the α;
Each coefficient A in the best-fit function
jObtain in the following manner: for coefficient A
jIf, A
jAbsolute value less than a threshold value beta=1E-3, A is removed in test so
j* X
jAfter, match similarity R
2Can or can not reduce threshold gamma=5%,, just remove A if it is so much can not to reduce γ
j* X
jItem is promptly A
jAssignment is 0.
9. a kind of character recognition method based on fitting function according to claim 1 is characterized in that the highest template of the described searching similarity of step (4) is as follows as the method for recognition result:
(a) each best-fit vector that splits out according to figure is searched the stroke that mates most in the stroke set of knowledge base, generate symbolic vector;
(b) utilize symbolic vector in the glossary of symbols of knowledge base, to search the symbol of coupling, if stroke be orderly so in search procedure stroke and order all to mate; Search according to the quantity and the type of stroke so if stroke is unordered, do not consider to mate in proper order;
(c) if under orderly with the unordered situation of stroke, can not find stroke number identical can not find used stroke identical, then use near matching principle, look for immediate symbol:, then get each and can match the maximum character of correct stroke quantity in order for the orderly character vector of stroke; If stroke is unordered, then get and can match the maximum character of used stroke in all strokes, then as required, can take out the image of character vector, compare the affirmation result with image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010100179333A CN101763516B (en) | 2010-01-15 | 2010-01-15 | Character recognition method based on fitting functions |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010100179333A CN101763516B (en) | 2010-01-15 | 2010-01-15 | Character recognition method based on fitting functions |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101763516A true CN101763516A (en) | 2010-06-30 |
| CN101763516B CN101763516B (en) | 2012-02-29 |
Family
ID=42494672
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010100179333A Active CN101763516B (en) | 2010-01-15 | 2010-01-15 | Character recognition method based on fitting functions |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101763516B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184395A (en) * | 2011-06-08 | 2011-09-14 | 天津大学 | String-kernel-based hand-drawn sketch recognition method |
| CN102737254A (en) * | 2012-06-15 | 2012-10-17 | 常州南京大学高新技术研究院 | Identification method of mark image |
| CN103106402A (en) * | 2013-03-07 | 2013-05-15 | 成都威路特软件科技有限公司 | Recognition method for license plate Chinese characters based on reconstruction |
| CN104008405A (en) * | 2013-02-26 | 2014-08-27 | 佳能株式会社 | Handwritten character recognition method and handwritten character recognition device |
| WO2017140233A1 (en) * | 2016-02-18 | 2017-08-24 | 腾讯科技（深圳）有限公司 | Text detection method and system, device and storage medium |
| CN108052925A (en) * | 2017-12-28 | 2018-05-18 | 江西高创保安服务技术有限公司 | A kind of cell personnel archives intelligent management |
| CN108288064A (en) * | 2017-01-09 | 2018-07-17 | 北京京东尚科信息技术有限公司 | Method and apparatus for generating picture |
| CN108345853A (en) * | 2018-02-08 | 2018-07-31 | 吉首大学 | Character identifying method, device based on isomorphism theory and terminal device |
| CN108668129A (en) * | 2018-03-14 | 2018-10-16 | 西安万像电子科技有限公司 | A kind of coding method and device |
| CN108877768A (en) * | 2018-05-21 | 2018-11-23 | 广东省电信规划设计院有限公司 | Base prompts voice recognition method, device and computer equipment |
| CN109101973A (en) * | 2018-08-06 | 2018-12-28 | 掌阅科技股份有限公司 | Character recognition method, electronic equipment, storage medium |
| CN109359529A (en) * | 2018-09-11 | 2019-02-19 | 北京语言大学 | The treating method and apparatus of long pen section |
| WO2020147140A1 (en) * | 2019-01-17 | 2020-07-23 | 北京悦时网络科技发展有限公司 | Phrase code generation method and apparatus, phrase code recognition method and apparatus, and storage medium |
| CN113361368A (en) * | 2021-06-01 | 2021-09-07 | 福建福昕软件开发股份有限公司 | Method for identifying multiple types of lines in PDF (Portable document Format) through function fitting based on desktop software |
| CN114419740A (en) * | 2022-01-11 | 2022-04-29 | 平安普惠企业管理有限公司 | Living body detection method, device and equipment based on artificial intelligence and storage medium |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP3868637B2 (en) * | 1998-07-01 | 2007-01-17 | 株式会社リコー | Character recognition device, dictionary creation method, and recording medium |
| SE514377C2 (en) * | 1998-08-26 | 2001-02-19 | Gunnar Sparr | character recognition |
-
2010
- 2010-01-15 CN CN2010100179333A patent/CN101763516B/en active Active
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102184395A (en) * | 2011-06-08 | 2011-09-14 | 天津大学 | String-kernel-based hand-drawn sketch recognition method |
| CN102184395B (en) * | 2011-06-08 | 2012-12-19 | 天津大学 | String-kernel-based hand-drawn sketch recognition method |
| CN102737254A (en) * | 2012-06-15 | 2012-10-17 | 常州南京大学高新技术研究院 | Identification method of mark image |
| CN104008405A (en) * | 2013-02-26 | 2014-08-27 | 佳能株式会社 | Handwritten character recognition method and handwritten character recognition device |
| CN103106402A (en) * | 2013-03-07 | 2013-05-15 | 成都威路特软件科技有限公司 | Recognition method for license plate Chinese characters based on reconstruction |
| CN103106402B (en) * | 2013-03-07 | 2016-04-13 | 成都威路特软件科技有限公司 | A kind of license plate Chinese character recognition method based on rebuilding |
| WO2017140233A1 (en) * | 2016-02-18 | 2017-08-24 | 腾讯科技（深圳）有限公司 | Text detection method and system, device and storage medium |
| CN108288064A (en) * | 2017-01-09 | 2018-07-17 | 北京京东尚科信息技术有限公司 | Method and apparatus for generating picture |
| CN108052925A (en) * | 2017-12-28 | 2018-05-18 | 江西高创保安服务技术有限公司 | A kind of cell personnel archives intelligent management |
| CN108052925B (en) * | 2017-12-28 | 2021-08-03 | 江西高创保安服务技术有限公司 | Intelligent management method for community personnel files |
| CN108345853B (en) * | 2018-02-08 | 2020-10-23 | 吉首大学 | Character recognition method and device based on isomorphic theory and terminal equipment |
| CN108345853A (en) * | 2018-02-08 | 2018-07-31 | 吉首大学 | Character identifying method, device based on isomorphism theory and terminal device |
| CN108668129A (en) * | 2018-03-14 | 2018-10-16 | 西安万像电子科技有限公司 | A kind of coding method and device |
| CN108877768A (en) * | 2018-05-21 | 2018-11-23 | 广东省电信规划设计院有限公司 | Base prompts voice recognition method, device and computer equipment |
| CN108877768B (en) * | 2018-05-21 | 2020-12-11 | 广东省电信规划设计院有限公司 | Method and device for identifying stationary telephone prompt tone and computer equipment |
| CN109101973A (en) * | 2018-08-06 | 2018-12-28 | 掌阅科技股份有限公司 | Character recognition method, electronic equipment, storage medium |
| CN109359529A (en) * | 2018-09-11 | 2019-02-19 | 北京语言大学 | The treating method and apparatus of long pen section |
| WO2020147140A1 (en) * | 2019-01-17 | 2020-07-23 | 北京悦时网络科技发展有限公司 | Phrase code generation method and apparatus, phrase code recognition method and apparatus, and storage medium |
| US11334780B2 (en) | 2019-01-17 | 2022-05-17 | Yueshi Network Technology Development Co., Ltd. | Method for generating word code, method and device for recognizing codes |
| CN113361368A (en) * | 2021-06-01 | 2021-09-07 | 福建福昕软件开发股份有限公司 | Method for identifying multiple types of lines in PDF (Portable document Format) through function fitting based on desktop software |
| CN113361368B (en) * | 2021-06-01 | 2024-03-19 | 福建福昕软件开发股份有限公司 | Method for identifying multiple types of lines in PDF (Portable document Format) through function fitting based on desktop software |
| CN114419740A (en) * | 2022-01-11 | 2022-04-29 | 平安普惠企业管理有限公司 | Living body detection method, device and equipment based on artificial intelligence and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101763516B (en) | 2012-02-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101763516B (en) | Character recognition method based on fitting functions | |
| CN104517112B (en) | A kind of Table recognition method and system | |
| US20200065601A1 (en) | Method and system for transforming handwritten text to digital ink | |
| Harouni et al. | Online Persian/Arabic script classification without contextual information | |
| US8175389B2 (en) | Recognizing handwritten words | |
| US8295600B2 (en) | Image document processing device, image document processing method, program, and storage medium | |
| Arif et al. | Table detection in document images using foreground and background features | |
| Clausner et al. | Scenario driven in-depth performance evaluation of document layout analysis methods | |
| CN105260751B (en) | A kind of character recognition method and its system | |
| Agnihotri | Offline handwritten Devanagari script recognition | |
| CN105139041A (en) | Method and device for recognizing languages based on image | |
| JP2001325564A (en) | Retrieval of handwritten comment in chinese character in cursive style based upon radical model | |
| CN102194123A (en) | Method and device for defining table template | |
| CN105117740A (en) | Font identification method and device | |
| CN109857912A (en) | A kind of font recognition methods, electronic equipment and storage medium | |
| Azeem et al. | Online arabic handwritten digits recognition | |
| CN105184329A (en) | Cloud-platform-based off-line handwriting recognition method | |
| CN113657274A (en) | Table generation method and device, electronic equipment, storage medium and product | |
| CN104834891A (en) | Method and system for filtering Chinese character image type spam | |
| Obaidullah et al. | Structural feature based approach for script identification from printed Indian document | |
| Inunganbi et al. | Recognition of handwritten Meitei Mayek script based on texture feature | |
| Lamghari et al. | Template matching for recognition of handwritten Arabic characters using structural characteristics and Freeman code | |
| CN111241329A (en) | Image retrieval-based ancient character interpretation method and device | |
| Tang et al. | Online chemical symbol recognition for handwritten chemical expression recognition | |
| Firdaus et al. | Arabic letter segmentation using modified connected component labeling |
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 | ||
| C41 | Transfer of patent application or patent right or utility model | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20160414 Address after: 210042 Xu Zhuang Software Park, 699-27 Xuanwu Avenue, Xuanwu District, Jiangsu, Nanjing Province, Room 204, South Building, F, three District, China Patentee after: JIANGSU MID-CENT HAIFU NETWORK TECHNOLOGY CO.,LTD. Address before: Yudaojie Baixia District of Nanjing City, Jiangsu Province, No. 29 210016 Patentee before: Nanjing University of Aeronautics and Astronautics |
|
| TR01 | Transfer of patent right |
Effective date of registration: 20231123 Address after: Room 751, Yingying Building, 99 Tuanjie Road, Yanchuangyuan, Jiangbei New District, Nanjing, Jiangsu Province, 211899 Patentee after: Nanjing Ruishengda Technology Co.,Ltd. Address before: Room 204, South Building F, R&D Zone 3, Xuzhuang Software Park, No. 699-27 Xuanwu Avenue, Xuanwu District, Nanjing City, Jiangsu Province, 210042 Patentee before: JIANGSU MID-CENT HAIFU NETWORK TECHNOLOGY CO.,LTD. |
|
| TR01 | Transfer of patent right | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240328 Address after: Room 158, No. 2 Huayuan Road, Xuanwu District, Nanjing City, Jiangsu Province, 210042 Patentee after: Shanxing (Nanjing) Network Engineering Co.,Ltd. Country or region after: China Address before: Room 751, Yingying Building, 99 Tuanjie Road, Yanchuangyuan, Jiangbei New District, Nanjing, Jiangsu Province, 211899 Patentee before: Nanjing Ruishengda Technology Co.,Ltd. Country or region before: China |
|
| TR01 | Transfer of patent right |