CN104899965B - A kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine - Google Patents
A kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine Download PDFInfo
- Publication number
- CN104899965B CN104899965B CN201510251098.2A CN201510251098A CN104899965B CN 104899965 B CN104899965 B CN 104899965B CN 201510251098 A CN201510251098 A CN 201510251098A CN 104899965 B CN104899965 B CN 104899965B
- Authority
- CN
- China
- Prior art keywords
- character
- binaryzation
- sequence number
- matrix
- subtemplate
- 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
- 238000000034 method Methods 0.000 title claims abstract description 61
- 239000011159 matrix material Substances 0.000 claims description 53
- 238000010606 normalization Methods 0.000 claims description 20
- 238000005070 sampling Methods 0.000 claims description 12
- 230000011218 segmentation Effects 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 6
- 238000006243 chemical reaction Methods 0.000 claims description 6
- HUTDUHSNJYTCAR-UHFFFAOYSA-N ancymidol Chemical compound C1=CC(OC)=CC=C1C(O)(C=1C=NC=NC=1)C1CC1 HUTDUHSNJYTCAR-UHFFFAOYSA-N 0.000 claims description 5
- 239000000047 product Substances 0.000 description 14
- 238000010586 diagram Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000001154 acute effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000007795 chemical reaction product Substances 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000005574 cross-species transmission Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 229910052709 silver Inorganic materials 0.000 description 1
- 239000004332 silver Substances 0.000 description 1
Landscapes
- Character Input (AREA)
Abstract
The embodiment of the invention discloses a kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine, the method is identified by the matching template of character to character to be identified.After the image for obtaining paper money sequence number is converted into gray-scale map first, the identification region of character is determined by first positioning, binaryzation, repositioning, then pass through to cause that the area size of the character is in the same size with matching template under normalizing, then the character of matching is filtered out by dot product, complete identification.The method not only avoid the influence of noise, while filter step can be omitted；Suitable for current main flow various currency types including RMB, dollar, Euro, Hongkong dollar, yen etc..
Description
Technical field
The invention belongs to automatic identification technology field, particularly a kind of multinational paper money sequence number knowledge based on cleaning-sorting machine
Other method, relate to combine the secondary splitting method of bank note priori, and the character identifying method based on template matches.
Background technology
In financial field, each banknote has a unique sequence number, equivalent to the identity card of bank note, therefore sequence
Row number turns into the important mark of bank note.Either classify or false proof, sequence number all plays vital effect.And current one
The caseload of a little economic disputes and illegal Capital Flow increasingly increases, and also has strike violence bank note to plunder and wash dirty money in addition
Aspect, sequence number is even more sharp weapon of the state apparatus to banknote drifting management, is conducive to social stability.Current domestic a few large-scale silver of family
Capable branch and subbranch are equipped with middle-size and small-size cleaning-sorting machine mostly, add other bank outlets and ATM is equipped with, and the demand of cleaning-sorting machine exists
Just reach 100,000 within 2010, and in sustainable growth.Additionally, economic globalization is development trend, existing sorting
Machine is also growing for the identification function demand of foreign currency, it is therefore desirable to which the sequence number to various countries' bank note is identified, existing
Cleaning-sorting machine mainly realizes the sequence number identification to single currency type, limits range of application.
Current paper money sequence number identification mainly has three steps, image preprocessing, Character segmentation and character recognition.Existing
Have in technology, identification is main just for 2005 new edition RMB, and using narrow range, and new edition RMB background is single, sequence
Number identification difficulty it is relatively low.In the prior art secondary splitting method, first time dividing method have been used in Character segmentation step
It is to be calculated based on trip point, it is easy to printed and pollution effect, and it is very sensitive for the shading value of image；Second
The binaryzation of fixed threshold is divided into, record edge trip point is segmentation result, sensitive to light and shade, it is impossible to process background complexity
Bank note, such as dollar.Closest interpolation algorithm is used in character normalization, each pixel is required to recalculate gray scale,
The character more fixed for size this method reduce efficiency.In terms of identification, current method has two kinds, and method one is grid
Method, several grids (number of grid is 20-40) are divided into by character, and each grid has 20-30 pixel, i.e., each character will be located
Reason will at least travel through 500 pixels of statistics, although precision is higher, but sacrifices arithmetic speed significantly；Method two is more excellent, is base
The sorting technique of sequential machine is supported in SVM, its method is to set up 0-9 and 26 the 36 of numeral support sequential machine,
Error rate is relatively low, and in 2-3% or so, but processing speed is slow average at 50-70ms/, due to note input, rotation correction, intake sequence
Row number image needs the time of 40ms in itself, it is impossible to reach in the industry《GB 16999-2010 RMB-banknote discriminating device current techique bars
Part》Sequence number recognition speed requirement >=700 (/min) of A grades of cleaning-sorting machine of middle requirement, the standard of average handling time 80ms.
In addition, complicated for dollar bill background, the problems such as Euro number difference in size is very big, prior art is not all solved.Therefore pin
To drawbacks described above present in currently available technology, it is necessary to be studied in fact, to provide a kind of scheme, solves in the prior art
The defect of presence, it is to avoid cause paper money sequence number to recognize that currency type is single, recognition speed is low, accuracy rate problem not high.
The content of the invention
To solve the above problems, object of the present invention is to provide a kind of multinational paper money sequence number knowledge based on cleaning-sorting machine
Other method, using the secondary splitting method for combining bank note priori, and the character identifying method based on template matches, solution
Having determined during existing paper money sequence number is recognized can not be adapted to complex background and character boundary variation issue, while recognition speed is carried
It is high more than 8 times.
To achieve the above object, the technical scheme is that：A kind of multinational paper money sequence number identification based on cleaning-sorting machine
Method, it is characterised in that the method is：The image of paper money sequence number is obtained, and image is converted to gray-scale map I, then to ash
Degree figure I follows the steps below treatment：
Step 1：The bank note prior information of different currency types, face amount, version information is obtained, the prior information is in sequence number
The position distribution of each character；Positioned at the beginning of entering line character to gray-scale map using prior information, according to first positioning result to gray-scale map
Binary conversion treatment is carried out, will be repositioned after the image projection after binaryzation；According to repositioning result, to the figure after binaryzation
As being split, the repositioning binaryzation matrix of each character is obtained；
Step 2, the repositioning binaryzation matrix obtained to step 1 carries out character normalization, obtains normalizing binaryzation square
Battle array Cn, CnSize be A × B, the size with each subtemplate in matching template set M is identical；The character normalization method is
Arest neighbors interpolation method or bilinear interpolation method, arest neighbors interpolation method is used for the currency type that character shape is fixed；For character shape
The currency type of shape change uses bilinear interpolation method；
Step 3：By CnRespectively with matching template set M each match subtemplate and carry out dot product, and dot product is multiplied
Each element summation of product matrix, obtains element summation r, if CnSubtemplate M is matched with character KKDuring dot product, element summation r takes
Maximum is obtained, i.e.,A=1,2 ..., A, b=1,2 ..., B；Then K is identification
As a result；
Positioned at the beginning of character described in step 1, specifically include following steps：
(1) the original position m of the vertical direction of paper money sequence number character is determined*, m*=argmaxm(y·S)；
Wherein, S ∈ RpIt is gray-scale map I in the projection vector of vertical direction, I ∈ Rp×q, Rp×qRepresent the real number square of p rows q row
Battle array；S=[S1,S2,…,Sp]T,I=1,2 ..., p, SiIt is i-th element in projection vector S；Y is input picture
In the projection signal of vertical direction, y=[y1,y2,…,yp]；
yi=1+s (i-m-h)-s (i-m), i=1,2 ..., p, s (i) they are step signals,H is sequence
The height of highest character in number, value is determined by the prior information of bank note, is known terms；M ∈ [0, p-h] are sequence numbers perpendicular
Nogata to original position variable；
(2) the original position l of the horizontal direction of paper money sequence number character is determined*, l*=argmaxl(x·G)；
Wherein, G ∈ RqIt is input picture projection vector in the horizontal direction, G=[G1,G2,…,Gq]T,j
=1,2 ..., q, GjIt is j-th element in projection vector G；X is input picture projection signal in the horizontal direction；X=[x1,
x2,…,xq]；xj=1+s (j-l-d1-w1)-s(j-l-d1)+s(j-l-d2-w2)-s(j-l-d2)+…+s(j-l-dk-wk)-s
(j-l-dk), j=1,2 ..., q, k are the number of sequence sign character, w1,w2,…,wkRespectively the 1,2nd ..., the k width of character
Degree；The lower left corner of first character is as origin with sequence number, d1,d2,…,dkIt is each character lower-left Angle Position on sequence number
Relative to the distance of this origin, d1=0；k、w1,w2,…,wkAnd d1,d2,…,dkCan determine to take by the prior information of bank note
Value；L ∈ [0, q-w] are sequence number original position variables in the horizontal direction；
(3) according to the first position location (l of first character*,m*)、d1,d2,…,dkAnd w1,w2,…,wkObtain sequence number
The first positioning result of upper other characters, specially：
If the first position location of n-th character is：Horizontal direction (xn1,xn2), vertical direction (yn1,yn2)；Wherein xn1It is
The original position of the character horizontal direction, xn1=l*+dn；xn2It is the final position of the character horizontal direction, xn2=xn1+wn；yn1
It is the original position of the character vertical direction, yn1=m*；yn2It is the final position of the character vertical direction, yn2=m*+h。
Further, the binary conversion treatment described in step 1, specifically includes following steps：
(1) binaryzation region is determined, the binaryzation region is the horizontal original position x of first character11To k-th word
The horizontal end position x of symbolk2, the vertical original position y of first character11To the vertical final position y of first character12, i.e.,
It is (x that binaryzation level is interval11,xk2), it is (y that binaryzation is interval vertically11,y12)；
(2) binaryzation is carried out to the character in binaryzation region, obtains the first positioning binaryzation matrix of each character, if the
The n first positioning binaryzation matrix of character is Nn, its size is (yn2-yn1)×(xn2-xn1)；The threshold value that binaryzation is used is led to
Two-peak method is crossed to be calculated.
Further, the character repositioning described in step 1, specifically includes following steps：
(1) according to the first positioning binaryzation matrix N of characternVertical direction projection is carried out to each character, vertical side is obtained
To discrete series, the vertical direction discrete series for such as being obtained to n-th character is Wn；To WnTraveled through, met continuous three
Individual element is all higher than given threshold tn1Or an element is more than given threshold tn2First element position, as the perpendicular of character
The starting position coordinates of straight repositioning, are designated as yn1', the position of last element of one of two above condition is met as perpendicular
The final position coordinate of straight repositioning, is designated as yn2′；Wherein tn1=0.5wn, tn2=wn-3；According to (yn1′,yn2') update each
The binaryzation matrix of character, obtains new binaryzation matrix Nn', size is (yn2′-yn1′)×(xn2-xn1)；
(2) according to new binaryzation matrix Nn' horizontal direction projection is carried out to each character, obtain the discrete of horizontal direction
Sequence, the horizontal direction discrete series for such as being obtained to n-th character is Ln；To LnTraveled through, meeting, continuous three elements are equal
More than given threshold tn1' or element be more than given threshold tn2' first element position, the level as character is fixed again
The starting position coordinates of position, are designated as xn1', the position for meeting last element of one of two above condition is fixed again as level
The final position coordinate of position, is designated as xn2′；Wherein tn1(the y of '=0.5n2′-yn1'), tn2'=(yn2′-yn1′)-3；According to (xn1′,
xn2') the binaryzation matrix of each character is updated, obtain repositioning binaryzation matrix Nn", size is (yn2′-yn1′)×(xn2′-
xn1′)。
Further, the M of matching template set described in step 3 is obtained in the following manner：
(1) sampling obtains the U normalization binaryzation matrix D of certain character KK1,DK2,…,DKU, this U normalization two-value
Change diverse location of the matrix sampling from different paper money sequence number or identical paper money sequence number；
Counting statistics matrixU=1,2 ..., U, AKIn value minimum 0, be U to the maximum；Built based on this
Vertical preliminary subtemplate MK', specially：To AKThe point (i ', j ') of (i ', j ') >=0.8U, i.e. dense parts, MK' (i ', j ') take
10；To AKThe point (i ', j ') of (i ', j ')≤0.2U, i.e., sparse part, MK' (i ', j ') take -10；To 0.2U<AK(i′,j′)<
The point (i ', j ') of 0.8U, MK' (i ', j ') take 0；
Repeat the above steps, corresponding preliminary subtemplate is set up to different characters, obtain preliminary matching template
Set M'；
(2) preliminary matching template set M' is matched with the normalization binaryzation matrix of sampling in step (1)；Root
Matching template set is adjusted according to matching result, specially：
The normalization binaryzation matrix D of the sampling of character KKuIt is each with preliminary matching template set M'
Individual subtemplate carries out dot product, and each element summation of product matrix is obtained to dot product, obtains element summation r',A=1,2 ..., A, b=1,2 ..., B；K' is any character；If K'=
During K, MK' be character K matching subtemplate, MK=MK'；If K ≠ K', the subtemplate M for needing adjustment preliminaryK', method is：It is right
In preliminary subtemplate MK' and MK", for MK' (i ", j ")=MK′' (i ", the point of j ") (i ", j "), then MK(i ", j ")=MK'
(i″,j″)-1；For MK(i″,j″)≠MK′(i ", the point of j ") (i ", j "), then MK(i ", j ")=MK'(i″,j″)+1；So as to obtain
Obtain matching template MK；
Repeat the above steps, corresponding preliminary subtemplate, the matching mould after being adjusted are adjusted to different characters
Plate set.
The beneficial effects of the invention are as follows：
(1) make segmentation result more accurate using the secondary splitting of the prior information of number of paper money relative position distribution, and
And the influence of noise is avoided, while filter step can be omitted；
(2) in recognizing, the template of foundation reduces 150-180 pixel, and identification takes 1/10 of method before being, in addition
Accuracy is significantly improved compared with SVMs, and method is flexible, it is adaptable to which various currency types of current main flow include RMB, U.S.
Unit, Euro, Hongkong dollar, yen etc..
(3) compared with simple secondary splitting and connected domain, SVMs recognition methods that prior art is used, this
Invention will pretreatment as optional process, segmentation result is changed small, properer character by background light and shade, can process vertical setting of types, big
Small and character ratio has the character of acute variation；The character Euclidean distance of current maximum is reached in terms of feature extraction, significantly
Template size is reduced, computational complexity is reduced.
Brief description of the drawings
The step of Fig. 1 is the multinational paper money sequence number recognition methods based on cleaning-sorting machine of embodiment of the present invention flow chart；
Fig. 2 is the RMB being partitioned into of the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Serial number image；
Fig. 3 is the dollar sequence being partitioned into of the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Column number image；
Fig. 4 is Euro sequence being partitioned into of the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Column number image；
Fig. 5 is the Hongkong dollar sequence being partitioned into of the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Column number image；
Fig. 6 is the Hongkong dollar series number image in Fig. 5 by the segmentation result schematic diagram after secondary splitting；
Fig. 7 is the sparse portion of the character of the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Point, its sparse part is respectively upper, middle and lower；
Fig. 8 is fixed again for the character " 3 " of the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Position binaryzation Matrix C3Schematic diagram；
The character " 3 " that Fig. 9 sets up for the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Schematic diagram with template.
Specific embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Conversely, the present invention covers any replacement done in spirit and scope of the invention being defined by the claims, repaiies
Change, equivalent method and scheme.Further, in order that the public has a better understanding to the present invention, below to of the invention thin
It is detailed to describe some specific detail sections in section description.Part without these details for a person skilled in the art
Description can also completely understand the present invention.
With reference to Fig. 1, the step of show the multinational paper money sequence number recognition methods based on cleaning-sorting machine of the embodiment of the present invention
Flow chart.Obtain the image of paper money sequence number, and image be converted to gray-scale map I, as shown in Fig. 2 to 5, respectively RMB,
Dollar, Euro, the sequence number gray-scale map image of Hongkong dollar.Then treatment is followed the steps below to I：
Step 1：The bank note prior information of different currency types, face amount, version information is obtained, the prior information is in sequence number
The position distribution of each character；Positioned at the beginning of entering line character to gray-scale map using prior information, according to first positioning result to gray-scale map
Binary conversion treatment is carried out, will be repositioned after the image projection after binaryzation；According to repositioning result, to the figure after binaryzation
As being split, the repositioning binaryzation matrix of each character is obtained.It is illustrated in figure 8 the repositioning binaryzation square of character " 3 "
Battle array C3。
Step 2, the repositioning binaryzation matrix obtained to step 1 carries out character normalization, obtains normalizing binaryzation square
Battle array Cn, CnSize be a × b, a × b be matching template set M in each subtemplate size；Normalized purpose is：Will
The size adjustment of the gray-scale map of identification is into the same size with matching template.The character normalization method is arest neighbors interpolation method
Or bilinear interpolation method, use arest neighbors interpolation method for the currency type that character shape is fixed；For the currency type that character shape changes
Using bilinear interpolation method；The eigenmatrix C that normalization is obtainednLength and width in 12-18；
Step 3：By CnRespectively with matching template set M each match subtemplate and carry out dot product, and dot product is multiplied
Each element summation of product matrix, obtains element summation r, if CnSubtemplate M is matched with character KKDuring dot product, element summation r takes
Maximum is obtained, i.e.,A=1,2 ..., A, b=1,2 ..., B；Then K is identification
As a result；
In concrete application example, I can be 8 bit depth gray-scale maps, and size is optimal in the range of 64*256 to 64*320.
Positioned at the beginning of character described in step 1, specifically include following steps：
(1) the original position m of the vertical direction of paper money sequence number character is determined*, m*=argmaxm(y·S)；
Wherein, S ∈ RpIt is gray-scale map I in the projection vector of vertical direction, I ∈ Rp×q, Rp×qRepresent the real number square of p rows q row
Battle array；S=[S1,S2,…,Sp]T,I=1,2 ..., p, SiIt is i-th element in projection vector S；
In other application example, in order to ensure that data do not spill over and save operation time, each element of S is removed
Using 256 as end product, such as following formula,
Y is projection signal of the input picture in vertical direction, y=[y1,y2,…,yp]；yi=1+s (i-m-h)-s (i-
M), i=1,2 ..., p, s (i) are step signals,H is the height of highest character in sequence number, by bank note
Prior information determine value, be known terms；M ∈ [0, p-h] are original position variable of the sequence number in vertical direction.
(2) the original position l of the horizontal direction of paper money sequence number character is determined*, l*=argmaxl(x·G)；
Wherein, G ∈ RqIt is input picture projection vector in the horizontal direction, G=[G1,G2,…,Gq]T,j
=1,2 ..., q, GjIt is j-th element in projection vector G；X is input picture projection signal in the horizontal direction；X=[x1,
x2,…,xq]；xj=1+s (j-l-d1-w1)-s(j-l-d1)+s(j-l-d2-w2)-s(j-l-d2)+…+s(j-l-dk-wk)-s
(j-l-dk), j=1,2 ..., q, k are the number of sequence sign character, w1,w2,…,wkRespectively the 1,2nd ..., the k width of character
Degree；The lower left corner of first character is as origin with sequence number, d1,d2,…,dkIt is each character lower-left Angle Position on sequence number
Relative to the distance of this origin, d1=0；k、w1,w2,…,wkAnd d1,d2,…,dkCan determine to take by the prior information of bank note
Value；L ∈ [0, q-w] are sequence number original position variables in the horizontal direction；
(3) according to the first position location (l of first character*,m*)、d1,d2,…,dkAnd w1,w2,…,wkObtain sequence number
The first positioning result of upper other characters, specially：
If the first position location of n-th character is：Horizontal direction (xn1,xn2), vertical direction (yn1,yn2).Wherein xn1It is
The original position of the character horizontal direction, xn1=l*+dn；xn2It is the final position of the character horizontal direction, xn2=xn1+wn；yn1
It is the original position of the character vertical direction, yn1=m*；yn2It is the final position of the character vertical direction, yn2=m*+h。
Binary conversion treatment described in step 1, specifically includes following steps：
(1) binaryzation region is determined, the binaryzation region is the horizontal original position x of first character11To k-th word
The horizontal end position x of symbolk2, the vertical original position y of first character11To the vertical final position y of first character12, i.e.,
It is (x that binaryzation level is interval11,xk2), it is (y that binaryzation is interval vertically11,y12)；
(2) binaryzation is carried out to the character in binaryzation region, obtains the first positioning binaryzation matrix of each character, if the
The n first positioning binaryzation matrix of character is Nn, its size is (yn2-yn1)×(xn2-xn1)；The threshold value that binaryzation is used is led to
Two-peak method is crossed to be calculated.
Character repositioning described in step 1, specifically includes following steps：
(1) according to the first positioning binaryzation matrix N of characternVertical direction projection is carried out to each character, vertical side is obtained
To discrete series, the vertical direction discrete series for such as being obtained to n-th character is Wn；To WnTraveled through, met continuous three
Individual element is all higher than given threshold tn1Or an element is more than given threshold tn2First element position, as the perpendicular of character
The starting position coordinates of straight repositioning, are designated as yn1', the position of last element of one of two above condition is met as perpendicular
The final position coordinate of straight repositioning, is designated as yn2′；Wherein tn1=0.5wn, tn2=wn-3；According to (yn1′,yn2') update each
The binaryzation matrix of character, obtains new binaryzation matrix Nn', size is (yn2′-yn1′)×(xn2-xn1)；
(2) according to new binaryzation matrix Nn' horizontal direction projection is carried out to each character, obtain the discrete of horizontal direction
Sequence, the horizontal direction discrete series for such as being obtained to n-th character is Ln；To LnTraveled through, meeting, continuous three elements are equal
More than given threshold tn1' or element be more than given threshold tn2' first element position, the level as character is fixed again
The starting position coordinates of position, are designated as xn1', the position for meeting last element of one of two above condition is fixed again as level
The final position coordinate of position, is designated as xn2′；Wherein tn1(the y of '=0.5n2′-yn1'), tn2'=(yn2′-yn1′)-3；According to (xn1′,
xn2') the binaryzation matrix of each character is updated, obtain repositioning binaryzation matrix Nn", size is (yn2′-yn1′)×(xn2′-
xn1′).It is illustrated in figure 6 the repositioning of Hongkong dollar, i.e. secondary splitting result figure.
The M of matching template set described in step 3 is obtained in the following manner：
(1) sampling obtains the U normalization binaryzation matrix D of certain character KK1,DK2,…,DKU, this U normalization two-value
Change diverse location of the matrix sampling from different paper money sequence number or identical paper money sequence number；
Counting statistics matrixU=1,2 ..., U, AKIn value minimum 0, be U to the maximum；Built based on this
Vertical preliminary subtemplate MK', specially：To AKThe point (i ', j ') of (i ', j ') >=0.8U, i.e. dense parts, MK' (i ', j ') take
10；To AKThe point (i ', j ') of (i ', j ')≤0.2U, i.e., sparse part, as shown in fig. 7, MK' (i ', j ') take -10；To 0.2U<AK
(i′,j′)<The point (i ', j ') of 0.8U, MK' (i ', j ') take 0；
Repeat the above steps, corresponding preliminary subtemplate is set up to different characters, obtain preliminary matching template
Set M.Preliminary matching template set M contains 36 subtemplates, respectively 26 capitalization of ten numerals of 0-9 and A-Z
Letter.
(2) preliminary matching template set M is matched with the normalization binaryzation matrix of sampling in (1).According to
Matching template set M is adjusted with result.Specially：
The normalization binaryzation matrix D of the sampling of character KKuIn preliminary matching template set M'
Each subtemplate carries out dot product, and each element summation of product matrix is obtained to dot product, obtains element summation r',A=1,2 ..., A, b=1,2 ..., B；K' is any character；If K'=
During K, MK' be character K matching subtemplate, MK=MK'；If K ≠ K', the subtemplate M for needing adjustment preliminaryK', method is：It is right
In preliminary subtemplate MK' and MK′', for MK' (i ", j ")=MK" (i ", the point of j ") (i ", j "), then MK(i ", j ")=MK'
(i″,j″)-1；For MK(i″,j″)≠MK'(i ", the point of j ") (i ", j "), then MK(i ", j ")=MK'(i″,j″)+1；So as to obtain
Obtain matching template MK；
Repeat the above steps, corresponding preliminary subtemplate, the matching mould after being adjusted are adjusted to different characters
Plate set.
Repeat the above steps, corresponding preliminary subtemplate, the matching mould after being adjusted are adjusted to different characters
Plate set.
It is illustrated in figure 9 the final template after character " 3 " is adjusted.
Presently preferred embodiments of the present invention is the foregoing is only, is not intended to limit the invention, it is all in essence of the invention
Any modification, equivalent and improvement made within god and principle etc., should be included within the scope of the present invention.
Claims (4)
1. a kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine, it is characterised in that the method is：Obtain bank note sequence
Number image, and image is converted to gray-scale map I, treatment then is followed the steps below to gray-scale map I：
Step 1：Obtain the bank note prior information of different currency types, face amount, version information, the prior information be in sequence number each
The position distribution of character；Positioned at the beginning of entering line character to gray-scale map using prior information, gray-scale map is carried out according to first positioning result
Binary conversion treatment, will be repositioned after the image projection after binaryzation；According to repositioning result, the image after binaryzation is entered
Row segmentation, obtains the repositioning binaryzation matrix of each character；
Step 2, the repositioning binaryzation matrix obtained to step 1 carries out character normalization, obtains normalizing binaryzation Matrix Cn,
CnSize be A × B, the size with each subtemplate in matching template set M is identical；The character normalization method is nearest
Adjacent interpolation method or bilinear interpolation method, arest neighbors interpolation method is used for the currency type that character shape is fixed；Become for character shape
The currency type of change uses bilinear interpolation method；
Step 3：By CnRespectively with matching template set M each match subtemplate and carry out dot product, and product matrix is obtained to dot product
Each element summation, element summation r is obtained, if CnSubtemplate M is matched with character KKDuring dot product, element summation r obtains maximum
Value, i.e.,A=1,2 ..., A, b=1,2 ..., B；Then K is recognition result；
Positioned at the beginning of character described in step 1, specifically include following steps：
(1) the original position m of the vertical direction of paper money sequence number character is determined*, m*=argmaxm(y·S)；
Wherein, S ∈ RpIt is gray-scale map I in the projection vector of vertical direction, I ∈ Rp×q, Rp×qRepresent the real number matrix of p rows q row；S
=[S1,S2,…,Sp]T,I=1,2 ..., p, SiIt is i-th element in projection vector S；Y is input picture perpendicular
Nogata to projection signal, y=[y1,y2,…,yp]；
yi=1+s (i-m-h)-s (i-m), i=1,2 ..., p, s (i) they are step signals,During h is sequence number
The height of highest character, value is determined by the prior information of bank note, is known terms；M ∈ [0, p-h] are sequence numbers in side vertically
To original position variable；
(2) the original position l of the horizontal direction of paper money sequence number character is determined*, l*=argmaxl(x·G)；
Wherein, G ∈ RqIt is input picture projection vector in the horizontal direction, G=[G1,G2,…,Gq]T,J=1,
2 ..., q, GjIt is j-th element in projection vector G；X is input picture projection signal in the horizontal direction；X=[x1,x2,…,
xq]；xj=1+s (j-l-d1-w1)-s(j-l-d1)+s(j-l-d2-w2)-s(j-l-d2)+…+s(j-l-dk-wk)-s(j-l-
dk), j=1,2 ..., q, k are the number of sequence sign character, w1,w2,…,wkRespectively the 1,2nd ..., the k width of character；With
The lower left corner of first character is origin, d in sequence number1,d2,…,dkBe on sequence number each character lower-left Angle Position relative to
The distance of this origin, d1=0；k、w1,w2,…,wkAnd d1,d2,…,dkValue can be determined by the prior information of bank note；l∈
[0, q-w] is sequence number original position variable in the horizontal direction；
(3) according to the first position location (l of first character*,m*)、d1,d2,…,dkAnd w1,w2,…,wkObtain sequence number on its
The first positioning result of its character, specially：
If the first position location of n-th character is：Horizontal direction (xn1,xn2), vertical direction (yn1,yn2)；Wherein xn1It is the character
The original position of horizontal direction, xn1=l*+dn；xn2It is the final position of the character horizontal direction, xn2=xn1+wn；yn1It is the word
Accord with the original position of vertical direction, yn1=m*；yn2It is the final position of the character vertical direction, yn2=m*+h。
2. method according to claim 1, it is characterised in that the binary conversion treatment described in step 1, specifically includes following
Step：
(1) binaryzation region is determined, the binaryzation region is the horizontal original position x of first character11To k-th character
Horizontal end position xk2, the vertical original position y of first character11To the vertical final position y of first character12, i.e. two-value
It is (x that change level is interval11,xk2), it is (y that binaryzation is interval vertically11,y12)；
(2) binaryzation is carried out to the character in binaryzation region, obtains the first positioning binaryzation matrix of each character, if n-th
The first positioning binaryzation matrix of character is Nn, its size is (yn2-yn1)×(xn2-xn1)；The threshold value that binaryzation is used is by double
Peak method is calculated.
3. method according to claim 1, it is characterised in that the character repositioning described in step 1, specifically includes following
Step：
(1) according to the first positioning binaryzation matrix N of characternVertical direction projection is carried out to each character, obtain vertical direction from
Sequence is dissipated, the vertical direction discrete series for such as being obtained to n-th character is Wn；To WnTraveled through, met continuous three elements
It is all higher than given threshold tn1Or an element is more than given threshold tn2First element position, as the fixed again vertically of character
The starting position coordinates of position, are designated as yn1', the position of last element of one of two above condition is met as fixed again vertically
The final position coordinate of position, is designated as yn2′；Wherein tn1=0.5wn, tn2=wn-3；According to (yn1′,yn2') update each character
Binaryzation matrix, obtains new binaryzation matrix Nn', size is (yn2′-yn1′)×(xn2-xn1)；
(2) according to new binaryzation matrix Nn' horizontal direction projection is carried out to each character, the discrete series of horizontal direction is obtained,
The horizontal direction discrete series for such as being obtained to n-th character is Ln；To LnTraveled through, be all higher than continuous three elements are met
Given threshold tn1' or element be more than given threshold tn2' first element position, as character level reposition
Starting position coordinates, are designated as xn1', meet what the position of last element of one of two above condition repositioned as level
Final position coordinate, is designated as xn2′；Wherein tn1(the y of '=0.5n2′-yn1'), tn2'=(yn2′-yn1′)-3；According to (xn1′,xn2′)
The binaryzation matrix of each character is updated, obtains repositioning binaryzation matrix Nn", size is (yn2′-yn1′)×(xn2′-xn1′)。
4. method according to claim 1, it is characterised in that the M of matching template set described in step 3 is in the following manner
Obtain：
(1) sampling obtains the U normalization binaryzation matrix D of certain character KK1,DK2,…,DKU, this U normalization binaryzation square
Diverse location of the battle array sampling from different paper money sequence number or identical paper money sequence number；
Counting statistics matrixU=1,2 ..., U, AKIn value minimum 0, be U to the maximum；It is preliminary based on this foundation
Subtemplate MK', specially：To AKThe point (i ', j ') of (i ', j ') >=0.8U, i.e. dense parts, MK' (i ', j ') take 10；To AK
The point (i ', j ') of (i ', j ')≤0.2U, i.e., sparse part, MK' (i ', j ') take -10；To 0.2U<AK(i′,j′)<The point of 0.8U
(i ', j '), MK' (i ', j ') take 0；
Repeat the above steps, corresponding preliminary subtemplate is set up to different characters, obtain preliminary matching template set
M'；
(2) preliminary matching template set M' is matched with the normalization binaryzation matrix of sampling in step (1)；According to
Matching template set is adjusted with result, specially：
The normalization binaryzation matrix D of the sampling of character KKuWith in preliminary matching template set M' each
Subtemplate carries out dot product, and each element summation of product matrix is obtained to dot product, obtains element summation r',A=1,2 ..., A, b=1,2 ..., B；K' is any character；If K'=K
When, MK' be character K matching subtemplate, MK=MK'；If K ≠ K', the subtemplate M for needing adjustment preliminaryK', method is：It is right
In preliminary subtemplate MK' and MK′', for MK' (i ", j ")=MK′' (i ", the point of j ") (i ", j "), then MK(i ", j ")=MK'
(i″,j″)-1；For MK(i″,j″)≠MK'(i ", the point of j ") (i ", j "), then MK(i ", j ")=MK'(i″,j″)+1；So as to obtain
Obtain matching template MK；
Repeat the above steps, corresponding preliminary subtemplate, the matching template collection after being adjusted are adjusted to different characters
Close.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510251098.2A CN104899965B (en) | 2015-05-15 | 2015-05-15 | A kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510251098.2A CN104899965B (en) | 2015-05-15 | 2015-05-15 | A kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104899965A CN104899965A (en) | 2015-09-09 |
| CN104899965B true CN104899965B (en) | 2017-06-23 |
Family
ID=54032607
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510251098.2A Expired - Fee Related CN104899965B (en) | 2015-05-15 | 2015-05-15 | A kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104899965B (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105336035B (en) * | 2015-10-28 | 2019-02-01 | 深圳怡化电脑股份有限公司 | A kind of method and system of dirty crown word number image classification |
| CN106780960B (en) * | 2015-11-18 | 2020-01-14 | 深圳怡化电脑股份有限公司 | Method and system for identifying currency of Iran paper money |
| CN106780961B (en) * | 2015-11-23 | 2020-02-07 | 深圳怡化电脑股份有限公司 | Method and system for identifying face value of Iran paper money |
| US20170309105A1 (en) * | 2016-04-25 | 2017-10-26 | Leadot Innovation, Inc. | Method of Determining Currency and Denomination of an Inserted Bill in a Bill Acceptor Having a Single Slot and Related Device |
| CN106296969B (en) * | 2016-08-18 | 2019-04-12 | 深圳怡化电脑股份有限公司 | The recognition methods and system of bank note |
| CN106447905B (en) * | 2016-09-12 | 2019-04-09 | 深圳怡化电脑股份有限公司 | A kind of bank note currency type recognition methods and device |
| CN108665606B (en) * | 2017-03-30 | 2021-02-02 | 深圳怡化电脑股份有限公司 | Method and device for identifying information-oriented paper money |
| CN107688813A (en) * | 2017-09-24 | 2018-02-13 | 中国航空工业集团公司洛阳电光设备研究所 | A kind of character identifying method |
| CN107665538B (en) * | 2017-09-28 | 2020-08-18 | 深圳怡化电脑股份有限公司 | Paper currency classification method, device and equipment and readable storage medium |
| CN108831005B (en) * | 2018-04-16 | 2019-10-25 | 华中科技大学 | A kind of Euro version recognition methods and the system of the multiple features fusion based on image |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101923741A (en) * | 2010-08-11 | 2010-12-22 | 西安理工大学 | Paper currency number identification method based on currency detector |
| CN102194275A (en) * | 2010-03-15 | 2011-09-21 | 党力 | Automatic ticket checking method for train tickets |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH09212707A (en) * | 1996-01-30 | 1997-08-15 | Oki Electric Ind Co Ltd | Paper sheets, discriminating device |
| CN101329725B (en) * | 2008-07-30 | 2010-10-06 | 电子科技大学 | Method for dividing fingerprint image based on gradient projection and morphology |
| CN102222384B (en) * | 2011-05-27 | 2013-10-16 | 尤新革 | Analysis method of multispectral image of note |
| CN103456075B (en) * | 2013-09-06 | 2015-11-25 | 广州广电运通金融电子股份有限公司 | A kind of bill handling method and device |
-
2015
- 2015-05-15 CN CN201510251098.2A patent/CN104899965B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102194275A (en) * | 2010-03-15 | 2011-09-21 | 党力 | Automatic ticket checking method for train tickets |
| CN101923741A (en) * | 2010-08-11 | 2010-12-22 | 西安理工大学 | Paper currency number identification method based on currency detector |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104899965A (en) | 2015-09-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104899965B (en) | A kind of multinational paper money sequence number recognition methods based on cleaning-sorting machine | |
| CN105243730B (en) | Paper Currency Identification and system | |
| CN105303676B (en) | A kind of version recognition methods of bank note and system | |
| CN102542660B (en) | Bill anti-counterfeiting identification method based on bill watermark distribution characteristics | |
| CN102800148B (en) | RMB sequence number identification method | |
| CN101923741B (en) | Paper currency number identification method based on currency detector | |
| CN102542655B (en) | Note anti-counterfeiting discrimination method based on fiber personality characteristics | |
| CN106339707B (en) | A kind of gauge pointer image-recognizing method based on symmetric characteristics | |
| CN106845542A (en) | Paper money number intelligent identification Method based on DSP | |
| CN104658097B (en) | A kind of rmb paper currency denomination identifying method of Histogram Matching based on image | |
| CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
| CN104680130A (en) | Chinese character recognition method for identification cards | |
| CN106503694B (en) | Digit recognition method based on eight neighborhood feature | |
| CN102024144A (en) | Container number identification method | |
| CN105574531A (en) | Intersection point feature extraction based digital identification method | |
| CN104464079A (en) | Multi-currency-type and face value recognition method based on template feature points and topological structures of template feature points | |
| CN104298989A (en) | Counterfeit identifying method and counterfeit identifying system based on zebra crossing infrared image characteristics | |
| CN107622271A (en) | Handwriting text lines extracting method and system | |
| CN108681735A (en) | Optical character recognition method based on convolutional neural networks deep learning model | |
| CN106846354B (en) | A kind of Book Inventory method on the frame converted based on image segmentation and random hough | |
| CN103824373A (en) | Bill image sum classification method and system | |
| CN107122775A (en) | A kind of Android mobile phone identity card character identifying method of feature based matching | |
| CN103886309A (en) | Method for identifying dollar denominations through facial recognition | |
| CN107742357A (en) | A kind of recognition methods of paper money number and device | |
| CN107358718A (en) | A kind of crown word number identification method, device, equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170623 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |