CN102521911A - Identification method of crown word number (serial number) of bank note - Google Patents
Identification method of crown word number (serial number) of bank note Download PDFInfo
- Publication number
- CN102521911A CN102521911A CN2011104248229A CN201110424822A CN102521911A CN 102521911 A CN102521911 A CN 102521911A CN 2011104248229 A CN2011104248229 A CN 2011104248229A CN 201110424822 A CN201110424822 A CN 201110424822A CN 102521911 A CN102521911 A CN 102521911A
- Authority
- CN
- China
- Prior art keywords
- image
- font size
- skeleton
- hat font
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention provides an identification method of a crown word number (serial number) of a bank note. The method comprises the following steps that: a crown word number (serial number) zone is localized on an RMB image under visible lights; a crown word number (serial number) identification method based on wavelet transformation local die minimal value and an improved crown word number (serial number) identification method supporting a vector machine are respectively used to independently identify a crown word number (serial number); and a weight-based discrimination method is used to carry out combination and decision on the two identification methods, so that a final identification result is obtained. According to the method, an RMB crown word number (serial number) can be identified rapidly and stably, so that it is beneficial to carry out monitoring and management on an RMB flow direction.
Description
Technical field
The present invention relates to the recognition methods of a kind of novel banknote hat font size, belong to the cash inspecting machine field.
Background technology
Now, in the China that economy develops rapidly, the circulation of RMB amount is big, more than the Currency Type, make the monitoring of Renminbi and management work face the increasing pressure, and the crucial management that is to be preced with font size of the monitoring of Renminbi and management.So most crucial problem is how to discern the hat font size exactly.But cash inspecting machine in the market is not high for the discrimination of hat font size.Therefore, the banknote hat font size recognition methods that works out a kind of efficient maturation is particularly important, is directly connected to the safety and the order of finance.
Traditional hat font size recognition methods is mainly studied around single method; Ignored some essential characteristic that hat font size itself possesses, discrimination does not reach the requirement of the state compulsory standard (GB16999-2010 " Renminbi identifier general technical specifications ") of up-to-date promulgation.And tradition hat font size recognizer only limits to be preced with the cleaner banknote in font size zone, and the dirty banknote in hat font size zone often can't be identified.
Summary of the invention
The object of the present invention is to provide the recognition methods of a kind of novel banknote hat font size, this method can be discerned the hat font size of Renminbi in fast and stable ground, thus the monitoring and the management that help Renminbi to flow to.
Technical scheme of the present invention is: the recognition methods of banknote hat font size comprises: at first location hat font size is regional on the Renminbi image under visible light; Use respectively then based on the hat font size recognition methods of minimizing hat font size recognition methods of wavelet transformation localized mode and improved SVMs and independently discern the hat font size; Use method of discrimination that combination decision is carried out in aforementioned two kinds of recognition methodss at last, draw final recognition result based on weights.
Described method; The method in hat font size zone, location comprises on the Renminbi image: at first the canny operator with dual threshold carries out rim detection to obtain edge image to the regional image of hat font size; Afterwards for the marginal point of edge image; In original image, find corresponding point also to grow under certain condition as seed points; Utilize high threshold that image and edge image after seed points is grown are merged again, and use the image after low threshold value is combined to carry out denoising, promptly obtained the hat font size area image after the binaryzation.
Described method, the method in hat font size zone, location specifically may further comprise the steps on the Renminbi image:
1) use the canny operator of dual threshold to handle to hat font size area image, the edge image that obtains is designated as I
1, wherein high threshold is designated as W;
2) use Gaussian filter to carry out preliminary denoising to hat font size area image, the image that obtains is designated as I
2
3) image I on the edge of
1In, for each marginal point, at I
2In find the identical point in position, and note I
2In the minimum point of gray-scale value in 8 neighborhoods of these points, be defined as seed points;
4) image I on the edge of
1In find the point of gradient greater than W, simultaneously at I
2In find the identical point in position, note I
2In the maximum point and minimum point of gray-scale value in 8 neighborhoods of these points, carry out producing after this step two point sets: gray-scale value maximum point set H={h
1, h
2..., h
nAnd the minimum point set L={l of gray-scale value
1, l
2..., l
n, h wherein
iAnd l
iBe respectively the gray-scale value of the maximum point of gray-scale value and minimum point in 8 neighborhoods of these points, i value 1,2,3 ... N, n are the number of the point of maximum point of gray-scale value or minimum;
5) the mean value H of difference set of computations H and L
pAnd L
p
6) use H
pDo threshold value to I
2Carry out binaryzation, the image that obtains is designated as I
3
7) merge image I
3With edge image I
1, obtain image I
4
8) corresponding to image I
4Each corresponding seed points, utilize the region growing algorithm process, obtain image I
5
9) use L
pDo threshold value to I
2Carry out binaryzation, the image that obtains is designated as I
6
10) merge image I
5And I
6, obtain binary image I
7
11) to image I
7Carry out denoising, obtain the image of final binaryzation;
12) normalization prefix sign character to be identified.
Described method, recognition methods is based on the minimizing hat font size of wavelet transformation localized mode: the skeleton that at first extracts the hat font size through wavelet transformation localized mode minimal value; Further carry out denoising then, eliminate the burr on the skeleton; Extract the various features of skeleton at last, thus identification hat font size.
Described method, the step of extracting hat font size skeleton through wavelet transformation localized mode minimal value comprises:
1) selects suitable yardstick to carry out wavelet transformation for input picture, calculate the mould value of wavelet coefficient; Utilize the character of wavelet coefficient to carry out the target and background separation simultaneously;
2) the mould value of wavelet transformation is carried out many thresholdings, extract the initialized skeleton of hat font size;
3) to step 2) skeleton that obtains judges that if skeleton is single pixel, this point is exactly effective skeleton point so; Otherwise, go to step 1) with regard to carrying out wavelet transformation once more;
4) all skeleton points that obtain based on step 3) obtain final skeleton.
Described method is extracted the various features identification of skeleton and is preced with in the method for font size, and the characteristic of required extraction comprises:
1) skeleton image particular row or row laterally passes through number of times and vertically passes through number of times;
2) the non-zero section number of the transverse projection of skeleton image particular row or row and longitudinal projection and non-zero segment length;
3) number of skeleton image bifurcation;
4) number on skeleton diagram image circle and limit.
Described method, the step of using method of discrimination based on weights to draw final recognition result comprises:
1) measures discrimination respectively based on the hat font size recognition methods of minimizing hat font size recognition methods of wavelet transformation localized mode and improved SVMs;
2) for the higher employing high weight of discrimination in the said two kinds of methods of step 1), the low weights of the employing that discrimination is lower;
3) for the character of input, if the identification of said two kinds of methods obtains is same R as a result, net result is exactly R; If two kinds of result's differences that method obtains are respectively R
1And R
2, so according to the discrimination and the step 2 of step 1)) weights calculate, obtain final result.
The invention has the beneficial effects as follows:
1) adopts effective image segmentation algorithm, overcome hat font size zone and had of the influence of the situation of spot, strengthened the recognition capability that hat font size zone is had the banknote of spot, greatly improved the discrimination of hat font size discrimination.The present invention has obviously improved the effect of binaryzation, has greatly made things convenient for the subsequent treatment of hat font size identification, helps to monitor and manage the circulation of Renminbi.
2) characteristic of being preced with font size has been described all-sidedly and accurately.Because traditional method, has been ignored some essential characteristic that the hat font size possesses just to the single single mode identification method of feature selecting of hat font size.This paper adopt two kinds independently method hat font size different character is described, improved the accuracy of hat font size identification in itself.
Description of drawings
Fig. 1 is based on the minimizing hat font size of wavelet transformation localized mode recognition methods synoptic diagram.
Fig. 2 is the hat font size recognition methods synoptic diagram based on the image segmentation algorithm of dual threshold and improved SVMs.
Fig. 3 laterally passes through number of times and vertically passes through number of times calculating sketch map for the specific row or column of skeleton image.
Fig. 4 is that the number of skeleton image bifurcation is calculated synoptic diagram.
Fig. 5 is an idiographic flow block diagram of the present invention.
Embodiment
The present invention adds based on the manifold description of hat font size on tradition hat font size recognizer basis, adopts the characteristic of diverse ways portrayal hat font size, can improve the discrimination of hat font size significantly.
The present invention is a kind of combined recognising method based on minimizing skeleton of localized mode and improved SVM; The present invention at first locatees hat font size zone on the Renminbi image under visible light; Use two kinds of independences then and be preced with font size recognition methods (being respectively hat font size recognition methods) identification hat font size efficiently based on minimizing hat font size recognition methods of wavelet transformation localized mode and improved SVMs; Use method of discrimination to carry out combination decision at last, draw final recognition result based on weights.
The present invention has following three key problem in technology points:
1) utilizes dual threshold that former figure is carried out binary conversion treatment, strengthened the recognition capability that hat font size zone is had the banknote of spot greatly;
2) employing is discerned the hat font size based on the minimizing skeleton recognition methods of wavelet transformation localized mode with based on improved SVM (SVMs) recognition methods, has portrayed the characteristic of hat font size comprehensively;
3) adopt decision method based on weights to carry out combination decision and draw the final recognition result of hat font size, make hat font size recognition result more stable, can improve discrimination effectively.
Crucial part of the present invention is two kinds of independences and is preced with font size recognition methods and effective decision method efficiently, as follows:
One, based on the recognition methods of the minimizing hat font size of wavelet transformation localized mode
Utilize wavelet transformation localized mode minimal value to extract the skeleton (strip) of hat font size.The minimizing position of the localized mode of wavelet coefficient is independent of the wavelet scale conversion, and it can extract the central point (skeleton) of symmetrical edge contour.Because having portrayed, the wavelet transformation local minimum changes pixel the most slowly in the gray level image.These points have two types, and one type is the point set of background area, and another kind of is the point set that is positioned on target's center's line, and they are skeleton.In this patent, selecting the derivative of cubic B-spline function is wavelet function.
ψ(x)＝θ′(x) (1)
Wherein, x is the wavelet function independent variable, and θ (x) is a cubic B-spline function, and ψ (x) is a wavelet function.
The performing step of skeletal extraction is:
(1) selects suitable yardstick to carry out wavelet transformation for input picture, calculate the mould value of wavelet coefficient; Utilize the character of wavelet coefficient to carry out the target and background separation simultaneously.
(2) the mould value of wavelet transformation is carried out many thresholdings, extract the initialized skeleton of hat font size.
(3) if skeleton is single pixel, this point is exactly effective skeleton point so, otherwise just carries out wavelet transformation once more, and search mould minimal value is as effective skeleton point; Thereby obtain final skeleton.
Further carry out denoising, eliminate the burr on the skeleton, thereby obtain visual effect skeleton preferably.
For skeleton image (pixel of this patent acquiescence pixel value 0 is a background, and the pixel of pixel value 1 is a prospect), extract following characteristic:
(1) particular row or row laterally passes through number of times and vertically passes through number of times
Pass through the calculating of number of times: for a width of cloth bianry image, first pixel be 1 or pixel value become 1 calculating from 0 and pass through 1 time, otherwise not so, the number of times that passes through as 100110 is 2.
Like Fig. 3, get the center section (suppose to get 6 row) of " 8 " of skeletonizing, it pass through vertically that number of times adds up and be 14.
(2) the non-zero section number of the transverse projection of particular row or row and longitudinal projection and non-zero segment length
Being defined as of zero section, non-zero section: if M is the projection array of piece image or parts of images, its value be 0,0,4,25,22,0,0,0,55,56,0,0,0,0}.Definition is that 0 part is zero section continuously, is not that 0 part is the non-zero section continuously.As among the array M preceding two 0, middle 30,40 of back, promptly this projection array has 3 zero sections; Have two non-zero sections 4,25, and 22} with 55,56}, promptly this projection array has 2 non-zero sections.
Non-zero segment length L
nCalculate as follows: if n non-zero groups of a projection array is combined into M
1, M
2... .., M
i, M wherein
1Label in the corresponding array is m
1, M
iLabel in the corresponding array is m
i, n coordinate length L so
n=m
i-m
1+ 1.
Then the coordinate number of M is 3+1=4, first coordinate length L
1=5-3+1=3, second coordinate length L
2=10-9+1=2.
So if get the right half part of " 6 " behind the skeletonizing, its transverse projection non-zero section number is 2.First coordinate length is that 2, the second coordinate length are 11.
(3) number of skeleton image bifurcation.
Being defined as of bifurcation: for the image of a skeletonizing, from figure why not take up an official post with 2 points to seek pixel value along different directions be not 0 point, can both find same point through behind the limited number of time, just say that this point that finds is a bifurcation.
Like Fig. 4, the bifurcation number of " 6 " behind the skeletonizing is 1, and the bifurcation number of letter " H " is 2.
(4) number on skeleton image " circle " and " limit ".
Being defined as of " circle ": for the image of a skeletonizing; From any one foreground pixel point; Seek the foreground pixel point by a direction along skeleton, through still getting back to starting point after the limited number of time search, the zone that the path of then searching for surrounds is exactly " circle ".If image has bifurcation, seek respectively from bifurcation along two paths and be not 0 point, still can find this bifurcation through behind the limited number of time, then this image has two " circles ".
Being defined as of " limit ": for the image of a skeletonizing, from any one foreground pixel point, seeking the foreground pixel point by any direction along skeleton, is background dot through what find after the limited number of time search, claims that then this image has one " limit ".The number value on " limit " is 0 or 1, promptly has " limit " or does not have " limit ".
So for example the number of " circle " of " 0 " behind the skeletonizing is 1, the number on " limit " is 0; The number of " circle " of letter " B " is 2; The number on " limit " is 0, and the number of " circle " of " 9 " is 1, and the number on " limit " is 1; The number of " circle " of letter " C " is 0, and the number on " limit " is 1.
Through extracting 26 letters and 10 corresponding essential characteristic more than the numeral in 36 characters of hat font size, just can describe out their unique characteristics, thereby they are identified.
Two, based on the hat font size recognition methods of the image segmentation algorithm and the improved SVMs of dual threshold
Two key points of this recognition methods are following:
(1) based on the image segmentation algorithm of dual threshold
Image segmentation is the committed step of pattern-recognition.A kind of new and effective binarization method has been adopted in the pre-service of this patent recognition methods, can effectively eliminate the contaminated influence in hat font size zone with because the different influences that the binaryzation result is caused of banknote version.This method is at first carried out rim detection with the canny operator of dual threshold to the regional image of hat font size; Afterwards for the marginal point of edge image; In original image, find corresponding point also to grow under certain condition as seed points; Utilize high threshold that this image and edge image are merged again, and use the image after low threshold value is combined to carry out denoising, just can arrive the image after the binaryzation.
(2) improved SVMs
SVMs is a learning system of using the linear function hypothesis space at high-dimensional feature space, and it is that it can solve linear inseparable problem by a learning algorithm training from Optimum Theory, like hat font size identification problem.SVMs is mapped to a high-dimensional feature space through pre-determined Nonlinear Mapping with input vector matrix X, in this higher dimensional space, makes up the optimal classification lineoid then.Nonlinear Mapping specifically is reflected in SVMs and adopts certain kernel function, thereby avoids in high-dimensional feature space, carrying out complicated calculating.
Because the identification of hat font size is an inseparable problem of linearity.So when handling this problem, Duoed a Nonlinear Mapping link during than processing linear separability problem.
Suppose that this Nonlinear Mapping is:
x?→ψ(x) (2)
Wherein, x is the sample characteristics collection, and ψ (x) is an x correspondence mappings function.
Optimization aim is:
Wherein, x
iCorresponding to the feature set of i training sample, y
iThe corresponding type label of expression, ψ (x
i) expression x
iThe correspondence mappings function, w and b presentation class device model parameter.
This affined optimization problem can be converted into:
Wherein, the corresponding antithesis Lagrangian of L (a) expression optimization problem, α
i, α
jRepresent i, the corresponding Lagrange multiplier of a j sample, n representes sample number.
In (2) formula, the sample of luv space is mapped to very higher-dimension or even infinite dimensional feature space through after the nonlinear transformation.If directly in this higher dimensional space, seek largest interval classification lineoid,, to be very difficult or even imponderable on calculating like (4) formula.
Satisfying under the Mercer condition,, realizing K (x through using kernel function K
i, x
j)=ψ (x
i) ψ (x
j), that is to say the inner product of directly calculating higher dimensional space with the variable in former space.So just avoided complicated Nonlinear Mapping process, thereby realized carrying out the computing of higher dimensional space at lower dimensional space.
Therefore, formula (4) becomes
At last, the decision function of Nonlinear Vector machine becomes:
Wherein, x representes sample to be tested eigen collection, and f (x) is the prediction of treating the label of test sample book.
The present invention is based on the theory of above traditional SVMs,, adopt gaussian radial basis function (RBF) K (x in conjunction with the specific characteristic of the hat font size that extracts before this paper
i, x
j)=exp (|| x
i-x
j||
2/ σ
2) (σ representes the variance of gaussian radial basis function) as the kernel function of SVMs.SVM model with training is classified to characteristic vector data to be measured, identification hat font size.
Based on above two key points, the concrete operations flow process of this recognition methods is following:
(1) use the canny operator of dual threshold (wherein high threshold is designated as W) to handle to hat font size area image, the image that obtains is designated as I
1
(2) use Gaussian filter to carry out preliminary denoising to hat font size area image, the image that obtains is designated as I
2
(3) image I on the edge of
1In, for each marginal point, at I
2In find the identical point in position, and note the minimum point of gray-scale value in 8 neighborhoods of these points, be defined as seed points.
(4) image I on the edge of
1In find the point of gradient greater than W, simultaneously at I
2In find the identical point in position, note the maximum point and minimum point of gray-scale value in 8 neighborhoods of these points, carry out this step can produce two point set: H={h afterwards
1, h
2..., h
nAnd L={l
1, l
2..., l
n, h wherein
nAnd l
nIt is the set that the gray scale of the maximum point of gray-scale value and minimum point is formed in 8 neighborhoods of these points.
(5) the mean value H of set of computations H and L
pAnd L
p
(6) use H
pDo threshold value to I
2Carry out binaryzation, the image that obtains is designated as I
3
(7) merge image I
3With edge image I
1, obtain image I
4
(8) corresponding to image I
4Each corresponding seed points, utilize the region growing algorithm process, obtain image I
5
(9) use L
pDo threshold value to I
2Carry out binaryzation, the image that obtains is designated as I
6
(10) merge image I
5And I
6, obtain binary image I
7
(11) image is carried out denoising, obtain the image of final binaryzation.
(12) normalization prefix sign character to be identified.
(13) training of hat font size proper vector: in the binary image sample after the normalization of hat font size, select the sample training of 5 identical characters, calculate the sample feature:
A) the transverse projection value is designated as V
1
B) longitudinal projection's value is designated as V
2
C) laterally be divided into 3 parts, calculate the accounting of pixel 1 in the each several part, be designated as V
3
D) vertically be divided into 3 parts, calculate the accounting of pixel 1 in the each several part, be designated as V
4
E) laterally pass through number of times, be designated as V
5
F) vertically pass through number of times, be designated as V
6
Remember that 36 character characteristic of correspondence vector matrixs are V
n={ V
1, V
2, V
3, V
4, V
5, V
6, wherein n is 1～36, corresponding 36 characters.
(14) foundation of hat font size SVM model: utilize the eigenvectors matrix of 36 characters, utilize formula (5) to set up corresponding SVM model, each model has a type label, and totally 36, corresponding to 36 prefix sign characters.
(15) identification of hat font size: for the image of each input, calculate this eigenvectors matrix, utilize judgements of making a strategic decision of hat font size SVM model, font size is preced with in identification.
Three, adopt decision method to draw hat font size result based on weights
Final result carries out according to following flow process:
(1) measures the discrimination of two kinds of methods respectively.
(2) for the higher employing high weight of discrimination in two kinds of algorithms, the low weights of the employing that discrimination is lower.
(3) for the character of input, if the identification of two kinds of methods obtains is same R as a result, net result is exactly R.If two kinds of result's differences that method obtains are respectively R
1And R
2, according to the best preferred principle of discrimination, the method that discrimination is high is as final result so.
The present invention adopts the recognition methods based on skeleton, and is careful and portrayed the characteristic of hat font size essence all sidedly, adopts based on improved SVM recognition methods, accurately described the characteristic of hat font size others.Combine these two kinds of methods jointly the hat font size to be discerned then, and adopt method of discrimination that the recognition result that two kinds of methods draw is carried out ruling, can improve the discrimination of hat font size significantly based on weights.
Embodiment:
Be the concrete identifying of example explanation banknote with the 5th cover Renminbi 100 yuan notes below.At first collect Renminbi under the visible light and contain the image of being preced with the font size face, orient the zone of hat font size.Utilize hat font size zone as source images, be divided into two independent step and carry out.
First: at first adopt formula (1) that image is carried out wavelet transformation, obtain the mould of wavelet coefficient, image is carried out micronization processes, finally obtain being preced with the skeleton of font size image based on the mould minimal value of part.Extract the various features of skeleton afterwards, describe the hat font size.If the hat font size is normalized to 24*18.Be the characteristic that example is extracted it with hat font size " 9 " so: it is 1 that bifurcated is counted; The number of turns is 1, and the limit number is 1, middle 8 row vertically pass through adding up and being 24 of volumes; The transverse projection coordinate number of the left side 12 row is 4; Two projected lengths are respectively 12 and 2, and the transverse projection coordinate number of the right 12 row is 2, and projected length is 15.Through the characteristic of portrayal hat font size, can discern the hat font size.
Second: at first utilize the binarization method based on dual threshold that original image is carried out binary conversion treatment, obtain being preced with the bianry image in font size zone, cutting and normalization hat font size extracts 36 prefix sign characters of hat font size characteristic of correspondence vector matrix V afterwards then
nThen, the V corresponding according to 36 prefix sign characters
nUtilizing formula (5)---improved SVM training draws the SVM model of 36 prefix sign characters.To the hat font size of input, utilize formula (6) to discern at last.For example being preced with font size and being normalized to 24*18, is example with hat font size " 9 ", and intrinsic dimensionality is 90 dimensions.Select 5 of each prefix sign characters of non-9, totally 175, test sample book formed in 175 of 9 prefix sign characters, trains model, and wherein σ gets 33.
At last, test two kinds of discrimination and analytical calculations of being preced with font sizes respectively and obtain weights separately.Utilize decision method then, identify net result based on weights.
Claims (7)
1. banknote hat font size recognition methods is characterized in that comprising: at first location hat font size is regional on the Renminbi image under visible light; Use respectively then based on the hat font size recognition methods of minimizing hat font size recognition methods of wavelet transformation localized mode and improved SVMs and independently discern the hat font size; Use method of discrimination that combination decision is carried out in aforementioned two kinds of recognition methodss at last, draw final recognition result based on weights.
2. method according to claim 1; It is characterized in that; The method in hat font size zone, location comprises on the Renminbi image: at first the canny operator with dual threshold carries out rim detection to obtain edge image to the regional image of hat font size; For the marginal point of edge image, in original image, find corresponding point also to grow under certain condition afterwards, utilize high threshold that image and edge image after seed points is grown are merged again as seed points; And use the image after low threshold value is combined to carry out denoising, promptly obtained the hat font size area image after the binaryzation.
3. method according to claim 2 is characterized in that, the method in hat font size zone, location specifically may further comprise the steps on the Renminbi image:
1) use the canny operator of dual threshold to handle to hat font size area image, the edge image that obtains is designated as I
1, wherein high threshold is designated as W;
2) use Gaussian filter to carry out preliminary denoising to hat font size area image, the image that obtains is designated as I
2
3) image I on the edge of
1In, for each marginal point, at I
2In find the identical point in position, and note I
2In the minimum point of gray-scale value in 8 neighborhoods of these points, be defined as seed points;
4) image I on the edge of
1In find the point of gradient greater than W, simultaneously at I
2In find the identical point in position, note I
2In the maximum point and minimum point of gray-scale value in 8 neighborhoods of these points, carry out producing after this step two point sets: gray-scale value maximum point set H={h
1, h
2..., h
nAnd the minimum point set L={l of gray-scale value
1, l
2..., l
n, h wherein
iAnd l
iBe respectively the gray-scale value of the maximum point of gray-scale value and minimum point in 8 neighborhoods of these points, i value 1,2,3 ... N, n are the number of the point of maximum point of gray-scale value or minimum;
5) the mean value H of difference set of computations H and L
pAnd L
p
6) use H
pDo threshold value to I
2Carry out binaryzation, the image that obtains is designated as I
3
7) merge image I
3With edge image I
1, obtain image I
4
8) corresponding to image I
4Each corresponding seed points, utilize the region growing algorithm process, obtain image I
5
9) use L
pDo threshold value to I
2Carry out binaryzation, the image that obtains is designated as I
6
10) merge image I
5And I
6, obtain binary image I
7
11) to image I
7Carry out denoising, obtain the image of final binaryzation;
12) normalization prefix sign character to be identified.
4. method according to claim 1 is characterized in that, recognition methods is based on the minimizing hat font size of wavelet transformation localized mode: the skeleton that at first extracts the hat font size through wavelet transformation localized mode minimal value; Further carry out denoising then, eliminate the burr on the skeleton; Extract the various features of skeleton at last, thus identification hat font size.
5. method according to claim 4 is characterized in that, the step of extracting hat font size skeleton through wavelet transformation localized mode minimal value comprises:
1) selects suitable yardstick to carry out wavelet transformation for input picture, calculate the mould value of wavelet coefficient; Utilize the character of wavelet coefficient to carry out the target and background separation simultaneously;
2) the mould value of wavelet transformation is carried out many thresholdings, extract the initialized skeleton of hat font size;
3) to step 2) skeleton that obtains judges that if skeleton is single pixel, this point is exactly effective skeleton point so; Otherwise, go to step 1) with regard to carrying out wavelet transformation once more;
4) all skeleton points that obtain based on step 3) obtain final skeleton.
6. method according to claim 4 is characterized in that, extracts the various features identification of skeleton and is preced with in the method for font size, and the characteristic of required extraction comprises:
1) skeleton image particular row or row laterally passes through number of times and vertically passes through number of times;
2) the non-zero section number of the transverse projection of skeleton image particular row or row and longitudinal projection and non-zero segment length;
3) number of skeleton image bifurcation;
4) number on skeleton diagram image circle and limit.
7. method according to claim 1 is characterized in that, the step of using method of discrimination based on weights to draw final recognition result comprises:
1) measures discrimination respectively based on the hat font size recognition methods of minimizing hat font size recognition methods of wavelet transformation localized mode and improved SVMs;
2) for the higher employing high weight of discrimination in the said two kinds of methods of step 1), the low weights of the employing that discrimination is lower;
3) for the character of input, if the identification of said two kinds of methods obtains is same R as a result, net result is exactly R; If two kinds of result's differences that method obtains are respectively R
1And R
2, so according to the discrimination and the step 2 of step 1)) weights calculate, obtain final result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110424822.9A CN102521911B (en) | 2011-12-16 | 2011-12-16 | Identification method of crown word number (serial number) of bank note |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110424822.9A CN102521911B (en) | 2011-12-16 | 2011-12-16 | Identification method of crown word number (serial number) of bank note |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102521911A true CN102521911A (en) | 2012-06-27 |
| CN102521911B CN102521911B (en) | 2014-03-12 |
Family
ID=46292812
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110424822.9A Active CN102521911B (en) | 2011-12-16 | 2011-12-16 | Identification method of crown word number (serial number) of bank note |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102521911B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800148A (en) * | 2012-07-10 | 2012-11-28 | 中山大学 | RMB sequence number identification method |
| CN103440665A (en) * | 2013-09-13 | 2013-12-11 | 重庆大学 | Automatic segmentation method of knee joint cartilage image |
| CN104318238A (en) * | 2014-11-10 | 2015-01-28 | 广州御银科技股份有限公司 | Method for extracting crown word numbers from scanned banknote images in banknote detection module |
| CN105336035A (en) * | 2015-10-28 | 2016-02-17 | 深圳怡化电脑股份有限公司 | Smudged serial number image classification method and system |
| CN106355744A (en) * | 2016-08-24 | 2017-01-25 | 深圳怡化电脑股份有限公司 | Image identification method and device |
| CN106650714A (en) * | 2016-10-08 | 2017-05-10 | 迪堡金融设备有限公司 | Paper note serial number identification method and apparatus |
| CN106683264A (en) * | 2017-02-20 | 2017-05-17 | 深圳怡化电脑股份有限公司 | Identification method and device of stained crown size of RMB |
| CN106710063A (en) * | 2016-12-27 | 2017-05-24 | 陕西科技大学 | RMB serial number recognition method based on computer vision |
| CN106803258A (en) * | 2017-01-13 | 2017-06-06 | 深圳怡化电脑股份有限公司 | A kind of image processing method and device |
| CN106845469A (en) * | 2017-01-24 | 2017-06-13 | 深圳怡化电脑股份有限公司 | A kind of Paper Currency Identification and device |
| CN106952392A (en) * | 2017-03-24 | 2017-07-14 | 深圳怡化电脑股份有限公司 | A kind of crown word number discrimination determines method and device |
| CN107067533A (en) * | 2017-04-14 | 2017-08-18 | 深圳怡化电脑股份有限公司 | The method and device that a kind of bank note differentiates |
| CN107331030A (en) * | 2017-06-23 | 2017-11-07 | 深圳怡化电脑股份有限公司 | A kind of crown word number identification method, device, equipment and storage medium |
| CN108022363A (en) * | 2016-11-02 | 2018-05-11 | 深圳怡化电脑股份有限公司 | A kind of bank note towards recognition methods and device |
| CN108154596A (en) * | 2016-12-04 | 2018-06-12 | 湖南丰汇银佳科技股份有限公司 | A kind of double comb paper money discrimination method based on images match |
| CN108734669A (en) * | 2017-04-24 | 2018-11-02 | 南京理工大学 | Image denoising method based on wavelet transformation Wiener filtering and edge detection |
| CN109086766A (en) * | 2018-06-06 | 2018-12-25 | 南京华科和鼎信息科技有限公司 | A kind of multi threshold fusion crown word number extracting method based on integrogram |
| CN109345684A (en) * | 2018-07-11 | 2019-02-15 | 中南大学 | A kind of multinational paper money number recognition methods based on GMDH-SVM |
| CN109697443A (en) * | 2017-10-23 | 2019-04-30 | 深圳怡化电脑股份有限公司 | A kind of paper money number dividing method and splitting equipment |
| CN111046866A (en) * | 2019-12-13 | 2020-04-21 | 哈尔滨工程大学 | Method for detecting RMB crown word number region by combining CTPN and SVM |
| CN112307867A (en) * | 2020-03-03 | 2021-02-02 | 北京字节跳动网络技术有限公司 | Method and apparatus for outputting information |
| CN112581904A (en) * | 2019-09-30 | 2021-03-30 | 华中科技大学 | Moire compensation method for brightness gray scale image of OLED (organic light emitting diode) screen |
| CN113269919A (en) * | 2020-12-28 | 2021-08-17 | 深圳市怡化时代科技有限公司 | Banknote identifying method, banknote identifying device, medium and banknote identifying apparatus |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000057313A (en) * | 1998-08-05 | 2000-02-25 | Glory Ltd | Light reception data corrector |
| KR20060033945A (en) * | 2004-10-18 | 2006-04-21 | 이-뱅킹텍 | Identifying method of note |
| CN102169608A (en) * | 2011-05-27 | 2011-08-31 | 徐端全 | Design method of multispectral paper currency detector |
| CN102184591A (en) * | 2011-04-13 | 2011-09-14 | 张文峰 | Cash counter and paper money management system |
| CN102222384A (en) * | 2011-05-27 | 2011-10-19 | 尤新革 | Analysis method of multispectral image of note |
-
2011
- 2011-12-16 CN CN201110424822.9A patent/CN102521911B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2000057313A (en) * | 1998-08-05 | 2000-02-25 | Glory Ltd | Light reception data corrector |
| KR20060033945A (en) * | 2004-10-18 | 2006-04-21 | 이-뱅킹텍 | Identifying method of note |
| CN102184591A (en) * | 2011-04-13 | 2011-09-14 | 张文峰 | Cash counter and paper money management system |
| CN102169608A (en) * | 2011-05-27 | 2011-08-31 | 徐端全 | Design method of multispectral paper currency detector |
| CN102222384A (en) * | 2011-05-27 | 2011-10-19 | 尤新革 | Analysis method of multispectral image of note |
Cited By (31)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102800148A (en) * | 2012-07-10 | 2012-11-28 | 中山大学 | RMB sequence number identification method |
| CN103440665A (en) * | 2013-09-13 | 2013-12-11 | 重庆大学 | Automatic segmentation method of knee joint cartilage image |
| CN103440665B (en) * | 2013-09-13 | 2016-09-14 | 重庆大学 | Automatic segmentation method of knee joint cartilage image |
| CN104318238A (en) * | 2014-11-10 | 2015-01-28 | 广州御银科技股份有限公司 | Method for extracting crown word numbers from scanned banknote images in banknote detection module |
| CN105336035A (en) * | 2015-10-28 | 2016-02-17 | 深圳怡化电脑股份有限公司 | Smudged serial number image classification method and system |
| 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 |
| CN106355744B (en) * | 2016-08-24 | 2019-07-26 | 深圳怡化电脑股份有限公司 | A kind of recognition methods of Indonesian Rupiah value of money and device |
| CN106355744A (en) * | 2016-08-24 | 2017-01-25 | 深圳怡化电脑股份有限公司 | Image identification method and device |
| CN106650714A (en) * | 2016-10-08 | 2017-05-10 | 迪堡金融设备有限公司 | Paper note serial number identification method and apparatus |
| CN108022363A (en) * | 2016-11-02 | 2018-05-11 | 深圳怡化电脑股份有限公司 | A kind of bank note towards recognition methods and device |
| CN108154596A (en) * | 2016-12-04 | 2018-06-12 | 湖南丰汇银佳科技股份有限公司 | A kind of double comb paper money discrimination method based on images match |
| CN106710063A (en) * | 2016-12-27 | 2017-05-24 | 陕西科技大学 | RMB serial number recognition method based on computer vision |
| CN106803258A (en) * | 2017-01-13 | 2017-06-06 | 深圳怡化电脑股份有限公司 | A kind of image processing method and device |
| CN106845469A (en) * | 2017-01-24 | 2017-06-13 | 深圳怡化电脑股份有限公司 | A kind of Paper Currency Identification and device |
| CN106845469B (en) * | 2017-01-24 | 2020-08-18 | 深圳怡化电脑股份有限公司 | Paper money identification method and device |
| CN106683264A (en) * | 2017-02-20 | 2017-05-17 | 深圳怡化电脑股份有限公司 | Identification method and device of stained crown size of RMB |
| CN106952392B (en) * | 2017-03-24 | 2019-07-09 | 深圳怡化电脑股份有限公司 | A kind of crown word number discrimination determines method and device |
| CN106952392A (en) * | 2017-03-24 | 2017-07-14 | 深圳怡化电脑股份有限公司 | A kind of crown word number discrimination determines method and device |
| CN107067533A (en) * | 2017-04-14 | 2017-08-18 | 深圳怡化电脑股份有限公司 | The method and device that a kind of bank note differentiates |
| CN108734669A (en) * | 2017-04-24 | 2018-11-02 | 南京理工大学 | Image denoising method based on wavelet transformation Wiener filtering and edge detection |
| CN107331030A (en) * | 2017-06-23 | 2017-11-07 | 深圳怡化电脑股份有限公司 | A kind of crown word number identification method, device, equipment and storage medium |
| CN107331030B (en) * | 2017-06-23 | 2019-11-15 | 深圳怡化电脑股份有限公司 | A kind of crown word number identification method, device, equipment and storage medium |
| CN109697443A (en) * | 2017-10-23 | 2019-04-30 | 深圳怡化电脑股份有限公司 | A kind of paper money number dividing method and splitting equipment |
| CN109086766A (en) * | 2018-06-06 | 2018-12-25 | 南京华科和鼎信息科技有限公司 | A kind of multi threshold fusion crown word number extracting method based on integrogram |
| CN109086766B (en) * | 2018-06-06 | 2021-03-09 | 南京华科和鼎信息科技有限公司 | Multi-threshold fusion crown word number extraction method based on integral graph |
| CN109345684A (en) * | 2018-07-11 | 2019-02-15 | 中南大学 | A kind of multinational paper money number recognition methods based on GMDH-SVM |
| CN112581904A (en) * | 2019-09-30 | 2021-03-30 | 华中科技大学 | Moire compensation method for brightness gray scale image of OLED (organic light emitting diode) screen |
| CN111046866A (en) * | 2019-12-13 | 2020-04-21 | 哈尔滨工程大学 | Method for detecting RMB crown word number region by combining CTPN and SVM |
| CN111046866B (en) * | 2019-12-13 | 2023-04-18 | 哈尔滨工程大学 | Method for detecting RMB crown word number region by combining CTPN and SVM |
| CN112307867A (en) * | 2020-03-03 | 2021-02-02 | 北京字节跳动网络技术有限公司 | Method and apparatus for outputting information |
| CN113269919A (en) * | 2020-12-28 | 2021-08-17 | 深圳市怡化时代科技有限公司 | Banknote identifying method, banknote identifying device, medium and banknote identifying apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102521911B (en) | 2014-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102521911B (en) | Identification method of crown word number (serial number) of bank note | |
| CN104866862B (en) | A kind of method of belt steel surface area-type defect recognition classification | |
| CN104392463B (en) | Image salient region detection method based on joint sparse multi-scale fusion | |
| CN106023220A (en) | Vehicle exterior part image segmentation method based on deep learning | |
| CN101859382B (en) | License plate detection and identification method based on maximum stable extremal region | |
| Garz et al. | Binarization-free text line segmentation for historical documents based on interest point clustering | |
| CN104182763B (en) | A kind of floristics identifying system based on flower feature | |
| CN103996018B (en) | Face identification method based on 4DLBP | |
| CN110210362A (en) | A kind of method for traffic sign detection based on convolutional neural networks | |
| CN104766046A (en) | Detection and recognition algorithm conducted by means of traffic sign color and shape features | |
| CN104751187A (en) | Automatic meter-reading image recognition method | |
| CN103093240A (en) | Calligraphy character identifying method | |
| CN104598885A (en) | Method for detecting and locating text sign in street view image | |
| CN102521561B (en) | Face identification method on basis of multi-scale weber local features and hierarchical decision fusion | |
| CN101706873A (en) | Identification method and device of digital-class limitation marking | |
| CN104573707B (en) | A kind of license plate Chinese character recognition method based on multi-feature fusion | |
| CN103020971A (en) | Method for automatically segmenting target objects from images | |
| CN103761743A (en) | Solid wood floor surface defect detecting method based on image fusion and division | |
| CN102254196A (en) | Method for identifying handwritten Chinese character by virtue of computer | |
| CN103413145A (en) | Articulation point positioning method based on depth image | |
| CN102663401A (en) | Image characteristic extracting and describing method | |
| CN106845542A (en) | Paper money number intelligent identification Method based on DSP | |
| CN106203401B (en) | A kind of rapid detection method of lane line | |
| Majid et al. | An Integrated Deep Learning Framework for Fruits Diseases Classification. | |
| CN104616016A (en) | Global feature and local feature combined texture feature description method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C53 | Correction of patent of invention or patent application | ||
| CB03 | Change of inventor or designer information |
Inventor after: You Xinge Inventor after: Jiang Tianyu Inventor after: Peng Qinmu Inventor after: Liu Yanshen Inventor after: Zheng Fei Inventor after: Peng Chong Inventor before: You Xinge Inventor before: Peng Qinmu Inventor before: Jiang Tianyu Inventor before: Zheng Fei Inventor before: Peng Chong |
|
| COR | Change of bibliographic data |
Free format text: CORRECT: INVENTOR; FROM: YOU XINGE PENG QINMU JIANG TIANYU ZHENG FEI PENG CHONG TO: YOU XINGE JIANG TIANYU PENG QINMU LIU YANSHEN ZHENG FEI PENG CHONG |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |