CN101799868A - Machine vision inspection method for paper money - Google Patents
Machine vision inspection method for paper money Download PDFInfo
- Publication number
- CN101799868A CN101799868A CN201010103644A CN201010103644A CN101799868A CN 101799868 A CN101799868 A CN 101799868A CN 201010103644 A CN201010103644 A CN 201010103644A CN 201010103644 A CN201010103644 A CN 201010103644A CN 101799868 A CN101799868 A CN 101799868A
- Authority
- CN
- China
- Prior art keywords
- image
- texture
- edge
- folder
- sides
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to a machine vision inspection method for paper money, in which an image taking device is firstly utilized to read in an image, the image picture displays the thickness of the paper money, the direction of the thickness of the paper money is set to be x direction, the direction of the width of the paper money is perpendicular to the picture, the direction of the length of the paper money is set to be y direction, and then whether the waist strip retainer of bundled paper money is correct is inspected. The method is applied in vision inspection of bundled paper money, such as waist strip retainer and seal identification, facing slip information identification, money bundle face value identification and the like, and realizes automation and high inspection efficiency and accuracy.
Description
Technical field
This method belongs to the image recognition application process, specifically is image recognition is applied to the whole machine vision inspection method for paper money of tying in detecting of bank note.
Background technology
Machine vision replaces human eye to do measurement and judgement with machine exactly.Vision Builder for Automated Inspection is meant that by machine vision product (be image-pickup device, as CMOS and CCD) Target Transformation that will detect becomes digital quantity signal, these digital quantity signals send special-purpose image processing system (dividing embedded and the video card mode) again to, and image processing system is provided with the detection task according to the mission requirements that will detect.Control on-the-spot equipment action according to discrimination result then.
The characteristics of Vision Builder for Automated Inspection are to improve the product quality and the production line automation degree of producing.Especially be not suitable for the dangerous work environment of manual work or the occasion that human eye is difficult to meet the demands at some, the machine in normal service vision substitutes the artificial vision; Simultaneously in industrial processes in enormous quantities, low and precision is not high with artificial visual inspection product quality efficient, the automaticity that can enhance productivity greatly and produce with machine vision detection method.And machine vision is easy to realize information integration, is the basic technology that realizes computer integrated manufacturing system.Similar application as, tobacco business image acquisition recognition system mainly is made up of ccd video camera, lighting device, industrial computer and visual processes software, the realtime graphic of CCD high-speed camera picked-up tobacco leaf, and send into the graphics processing unit of electrical control cubicles.Image processing software detects automatically according to the feature of various settings, discerns normal tobacco leaf and foreign matter, and the discovery foreign matter is then incited somebody to action notification action device for information about.
The waist bar of bundled bank note whether block and seal image such as Fig. 1 of detecting, in general bank note is to prick a formation A with waist bar B earlier, is that a bundle is tied up with strapping C with 10 A again, detects at last.If the different bank note of denomination is mixed in A in a bundle, will think to occur " folder handle " situation.On each waist bar B, stamped signature (operator's name chapter) should be arranged also.
Connected domain is meant the set of being made up of some pixels, and the pixel in this set has following characteristic: the grey level of 1 all pixels all is less than or equal to the rank of connected domain; Pixel in the 2 same connected domains communicates in twos, promptly has a complete path that is made of the element of this set between any two pixels.
Summary of the invention
In order to be embodied as the vision-based detection of pack of paper coin, the detection automatization level of bundled bank note is improved, detection speed and precision improve, and the present invention proposes a kind of bank note card machine vision detection method, and concrete technical scheme is as follows:
A kind of machine vision inspection method for paper money reads in image with image-pickup device earlier, and what image frame showed is banknote thickness, and the banknote thickness direction is made as the x direction, and the width of paper money direction is perpendicular to picture, and the length direction of bank note is made as the y direction; Again to the waist bar card of bundled bank note whether correctly detecting, concrete steps comprise:
1) connected domain analysis: choose suitable judgement image earlier, and choose largest connected territory, i.e. target area; Judge tentatively according to largest connected territory whether image is correct, correctly then enter next step;
2) seek flex point in the target area the corner of pack of paper coin, if find four flex points, then enter next step;
3) judge folder: respectively edge by the evaluating objects zone or texture judge whether folder, if having folder, finish to judge that the prompting folder is withdrawing from; If the folder handle does not then enter next step;
4) judge many handles or few handle: obtain white waist bar information in the target area earlier; Judge whether that by analyzing the waist bar number is correct; If mistake, then prompting is miscounting mistake, withdrawing from; If correct, then finish to judge that prompting right withdraws from;
In the described step 3), comprise step:
3.1) edge in evaluating objects zone: for the image that obtains flex point, earlier image is carried out texture analysis, promptly the texture that the thickness of bank note constitutes on the x direction is analyzed; Move towards stack to what analyzing image texture obtained, judge that banknote stacks direction; Then the both sides image edge addition that is judged to be the correspondence that stacks direction; Again the image edge after the stack is analyzed, then determined to exist the folder handle;
3.2) step 3.1) judge that conclusion is not press from both sides handle, then jump out step 3), if folder handle, the then texture of further analysis image are arranged:
When if texture is straight line, step 3.1) folder that obtains the result for false; Texture is a straight line, and promptly image edge in the left and right sides adds up to straight line, do not protrude or recessed, about the phase side uneven owing to tie up out-of-flatness and cause;
If the texture in the x direction is a curve, and bending direction is identical with the direction of judging the folder handle, and promptly texture is bent downwardly and adjudicates when pressing from both sides the position below image, and folder is the result false; This judgement has been got rid of because bill is tied up the wrong conclusion that bending causes judgement folder handle.See Fig. 3.
If the texture in the x direction is a curve, and the direction of bending direction and judgement folder handle is opposite, and promptly texture is bent downwardly and adjudicates when pressing from both sides the position above image, and folder is the result true.See Fig. 4.
In the described step 1), it is dark forcing background earlier, by strengthening picture contrast and image binaryzation, is partitioned into image boundary;
Promptly come mode that the texture of image is followed the tracks of by texture tracked again, analyze image zones of different deformation extent, and adjust original image state to the distortion by image target area internal tangent direction; Described this tangential direction is the direction vertical with the gradient direction of image;
Described texture tracked concrete steps are, difference according to the bill bundle deformation texture degree that analyzes, the influence of crushed element is remedied, in general because the strength trend of tying up, the edge part branch on image both sides upwarps or has a downwarp, and length direction can shorten, and according to deformation extent the artificial compensation of shortening of part is got on.
In the described step 1, described correct image is to find a connected region that similarly meets the rectangle closure of general pattern size with all bill bundle samples of learning;
The process of choosing in largest connected territory: use homomorphic filtering and dynamic histogram thresholding to analyze the method for alternate analysis, accurately set display foreground and background gray level difference below 20; Judge largest connected territory by image binaryzation, and largest connected territory is corroded and expansion process, filter the edge burr; Under the state at the no image of judgement edge, use background value and the maximum edge of waist bar preserved to simulate the sample edge again.Because cross old bill binding together, color is darker, and is close with black background, can not analyze the image edge of bill bundle when graphical analysis, but the color of waist bar and background colour differ often, simulate bill bundle sample image edge again by the maximum edge that finds the waist bar.
Described step 2) in, the method for seeking flex point is that employing is promptly at first found two the longest points of distance on the edge with image 45 degree angle translation search plans, relocates out two remaining points;
For the banknote flange of 4 angles generations or the seal flange of center section generation, adjust the disturbing factor of individual generation of elimination by once corroding and expanding, and determine banknote flange, the seal flange position that the manual waist bar afterbody that twines leans out or center section produces by the image gray levels analysis once more, by direct removing parts of images, eliminate the influence that flange brings.
Described step 3.1) concrete steps are, addition is carried out to the projected length at image boundary place in banknote both sides of the edge in the image obtain an one-dimension array, again this one-dimension array is analyzed, judge according to its fluctuation range whether its edge has protrusion or recessed, and have only when protrude or recessed continuous part length greater than the both sides of the edge overall length 10% the time, just determine to exist folder.
In the described step 4), white waist bar information obtains by self-adaption binaryzation in the target area:
Because the gray level difference of waist bar or strapping and banknote itself, at first extract white waist bar and strapping position, specifically be to obtain with self-adaption binaryzation method, again binary image is carried out the connected domain analysis, there is gray difference according to waist bar and strapping and banknote itself, determines their position by this species diversity;
The condition that may occur again according to the position, the method for taking to travel through is estimated the situation of waist bar, and by the traversal back different evaluation of weight as a result, the situation of finally determining to tie up the waist bar;
Is number identification what obtain by the shadowing analysis between every handle; Under the condition of number deficiency, increase every thickness average value is more promptly judged about under each situation of 5, it is consistent that the average thickness on both sides is wanted, to avoid because the erroneous judgement that the two bundle banknote waist bars coincidences that the distortion of waist bar produces bring.
In the step 4), in identifying, use red channel to reach the purpose that implies the name chapter, with the shade between image normalization and the outstanding every bundle.Specifically, be for a set image, change the curve of certain passage separately, can cause colour cast, according to actual conditions, the name chapter on the waist bar is red, uses red channel, can eliminate the difference that the unknown chapter image in famous Zhanghe brings; By an implicit name chapter, with the shade between image normalization and the outstanding every bundle, a name Zhang Yanse will desalinate and be white in the image red channel, improve card accuracy of identification.
In the step 4), brand-new banknote stacks in strict accordance with last 5 times 5 cross methods, and because brand-new stiff thickness strict conformance;
After in algorithm identified, removing the use standard method of analysis, fresh money also increased connected domain area comparison process: owing to can produce the higher pixel of gray-scale value, so under each situation of 5 of both sides, whether the connected domain area of judging both sides gray-scale value upper zone equates, to promote the precision of test result.
This method is applied in the vision-based detection of bundled bank note, and for example waist bar card handle and seal identification, seal information Recognition and the identification of money bundle face amount etc. realize robotization, detection efficiency and precision height.
Description of drawings
Fig. 1 is that coin is tied up synoptic diagram;
Fig. 2 specifically judges the image synoptic diagram, wherein
Among the figure, E be flex point, G1 place one side be stack direction one side, G2 place one side is that the opposite side that stacks direction, the appearance profile D that F is texture, image are largest connected territory;
Fig. 3 is a step 3.2) in folder be the result false synoptic diagram;
Fig. 4 is a step 3.2) in folder be the result true synoptic diagram;
Fig. 5 is that folder is synoptic diagram;
Fig. 6 is the uneven synoptic diagram in left and right side;
Fig. 7 is up and down wave and seal flange synoptic diagram;
Fig. 8 is a left and right sides wave synoptic diagram;
Fig. 9 is that brand-new banknote is tied up synoptic diagram,
Among the figure, H is that white portion, J1 are that last 5 folded banknotes, J2 are for 5 folding banknotes down;
Figure 10 is the process flow diagram that programming realizes this method.
Embodiment
The invention will be further described below in conjunction with accompanying drawing and embodiment.
A kind of machine vision inspection method for paper money reads in image with image-pickup device earlier, and what image frame showed is banknote thickness, and the banknote thickness direction is made as the x direction, and the width of paper money direction is perpendicular to picture, and the length direction of bank note is made as the y direction; Again to the waist bar card of bundled bank note whether correctly detecting, step comprises:
1) connected domain analysis: choose suitable judgement image earlier, and choose largest connected territory, i.e. target area; Judge tentatively according to largest connected territory whether image is correct, correctly then enter next step;
Correct images is meant the connected region that finds similarly to meet the rectangle closure of general pattern size with all bill bundle samples of learning;
Be to use homomorphic filtering and dynamic histogram thresholding to analyze the method for alternate analysis to choosing of largest connected territory, accurately set display foreground and background gray level difference below 20.Judge largest connected territory by image binaryzation, and largest connected territory is corroded and expansion process, filter the edge burr.Under the state at the no image of judgement edge, use background value and the maximum edge of waist bar preserved to simulate the sample edge again.
2) seek flex point in the target area the corner of pack of paper coin, if find four flex points, then enter next step;
3) judge folder: respectively edge by the evaluating objects zone or texture judge whether folder, if having folder, finish to judge that the program prompts folder is withdrawing from; If the folder handle does not then enter next step;
4) judge many handles or few handle: obtain white waist bar information in the target area by self-adaption binaryzation; Judge whether that by analyzing the waist bar number is correct; If mistake, program prompts withdraws from miscounting mistake; If correct, then finish to judge that program prompts correctly withdraws from.
In the described step 3), comprise step:
3.1) edge in evaluating objects zone: for the image that obtains flex point, earlier image is carried out texture analysis, promptly the texture that the thickness of bank note constitutes on the x direction is analyzed; Move towards stack to what analyzing image texture obtained, judge that banknote stacks direction; Then the both sides image edge addition that is judged to be the correspondence that stacks direction; Again the image edge after the stack is analyzed, judgement whether exist protrude or recessed continuous part whether greater than 10%; If greater than, then determine to exist the folder handle;
Specifically: addition is carried out to the projected length at image boundary place in banknote both sides of the edge in the image obtain an one-dimension array, again this one-dimension array is analyzed, judge according to its fluctuation range whether its edge has protrusion or recessed, and have only when protrude or recessed continuous part length greater than the both sides of the edge overall length 10% the time, just determine to exist folder.
3.2) step 3.1) judge that conclusion is not press from both sides handle, then jump out step 3), if folder handle, the then texture of further analysis image are arranged:
When if texture is straight line, step 3.1) folder that obtains the result for false; Texture is a straight line, and promptly image edge in the left and right sides adds up to straight line, do not protrude or recessed, about the phase side uneven owing to tie up out-of-flatness and cause.
If the texture in the x direction is a curve, and bending direction is identical with the direction of judging the folder handle, and promptly texture is bent downwardly and adjudicates when pressing from both sides the position below image, and folder is the result false; This judgement has been got rid of because bill is tied up the wrong conclusion that bending causes judgement folder handle.
If the texture in the x direction is a curve, and the direction of bending direction and judgement folder handle is opposite, and promptly texture is bent downwardly and adjudicates when pressing from both sides the position above image, and folder is the result true.
For step 1),
It is dark forcing background, by strengthening picture contrast and image binaryzation, is partitioned into image boundary.Mode (promptly coming the texture of image is followed the tracks of) by texture tracked by image target area internal tangent direction (vertical) with the gradient direction of image, analyze image zones of different deformation extent, and (method of adjustment is the difference according to the bill bundle deformation texture degree that analyzes to the preceding state of distortion to adjust original image, the influence of crushed element is remedied, in general because the strength trend of tying up, the edge part branch on image both sides upwarps or has a downwarp, and length direction can shorten, and according to deformation extent the artificial compensation of shortening of part is got on).
The method of using homomorphic filtering and dynamic histogram thresholding to analyze alternate analysis is specific as follows:
A, homomorphic filtering: (x y) can regard as by two component combination and forms, promptly piece image f
f(x，y)＝i(x，y).r(x，y)
(x is luminance component (an incident component) y) to i, is the light intensity that incides on the scenery;
(x y) is reflecting component to r, is the light intensity that is subjected to the scenery reflection.
Concrete steps are as follows:
(1) taken the logarithm simultaneously in the both sides of following formula earlier, promptly
Inf(x，y)＝Ini(x，y)+Inr(x，y)
(2) Fourier transform is got on the following formula both sides,
F(u，v)＝I(u，v)+R(u，v)
(3) (u, (u v), can obtain v) to handle F with a frequency-domain function H
H(u，v)F(u，v)＝H(u，v)I(u，v)+H(u，v)R(u，v)
(4) inverse Fourier transform gets to spatial domain
Hff(x，y)＝hi(x，y)+hr(x，y)
As seen getting image after strengthening is to get two parts by corresponding luminance component and reflecting component to be formed by stacking.
(5) again index is got on the following formula both sides,
g(x，y)＝exp|hff(x，y)|＝exp|hi(x，y)|+exp|hr(x，y)|
Here, be called the homomorphic filtering function, it can act on respectively on luminance component and the reflecting component.
Piece image gets luminance component and characterizes with changing slowly usually, and reflecting component then tends to rapid conversion.So after taking the logarithm, image gets the main corresponding luminance component of low frequency part of Fourier transform, and the main corresponding reflecting component of HFS.Suitable selective filter function will produce different responses with HFS to the low frequency part in the Fourier transform.Result can make the dynamic range of pixel gray scale or picture contrast be enhanced.
B, dynamically histogram thresholding analysis alternate analysis: for a pure background image that does not have bill bundle sample, its histogram is the mountain peak pattern in strict accordance with normal distribution, for an image that common bill bundle sample is arranged, its histogram is to have two mountain peak patterns that do not overlap with the result of pure background histogram stack, than the largest connected territory that is easier to just distinguish bill bundle image.But for old bill bundle, because its color is deep, very approaching with background color, its histogram is two patterns that the mountain peak intersects with the result of the histogrammic stack of pure background, can not find its largest connected territory with common connected domain analytical approach, so that intersection concave point that just according to circumstances adopts two mountain peaks to intersect is followed the threshold value of background as distinguishing sample.Because external environment condition is different, this concave point of image that different conditions is taken down also can be the value of a dynamic change, so for each pictures, all will get the threshold value of the point of this dynamic change as judgement sample and background.
Under the state at the no image of judgement edge, use background value and the maximum edge of waist bar preserved to simulate the sample edge again, because cross old bill binding together, color is darker, close with black background, when graphical analysis, can not analyze the image edge of bill bundle, but the color of waist bar and background colour differ often, simulate bill bundle sample image edge again by the maximum edge that finds the waist bar.
For step 2)
Used two independent positioning methods when this method is abandoned generally seeking the image border adopt with image 45 degree angle translation search plans (step of this search plan is at first to find two the longest points of distance on the edge, relocates out two remaining points).When this programme produces disappearance or distortion 4 jiaos of rectangles, produce fabulous search effect.
4 fixed points by this method positioning image, for the banknote flange of 4 angles generations or the seal flange of center section generation, adjust the disturbing factor of individual generation of elimination by once corroding and expanding, and determine banknote flange, the seal flange position that the manual waist bar afterbody that twines leans out or center section produces by the image gray levels analysis once more, by direct removing parts of images, eliminate the influence that flange brings.
Corrosion and expansion:
Obtain S behind the bar structure element S translation x
x, if S
xBe contained in X, we write down this x point, and all set of satisfying the x point composition of above-mentioned condition are called X by the result of S corrosion (Erosion).Corroding method is, takes the initial point of S and the point on the X to contrast singly, if the institute on the S has a few all in the scope of X, and the then point of the initial point correspondence of S reservation, otherwise this point is removed.Obtain S behind the bar structure element S translation x
x, if S
xIt is not empty intersecting with X, and we write down this x point, and all set of satisfying the x point composition of above-mentioned condition are called X by the result of S expansion (dilation).The method that expands is, takes the initial point of S and the point on the X to contrast singly, if there is a point to drop in the scope of X on the S, then the point of the initial point correspondence of S is an image just.
Image gray levels is analyzed:
Because can know according to graphical analysis, the texture color of bill bundle is saturate than waist bar, seal and banknote flange always, so utilize the otherness of this gray scale, this example is adding 80 standards as judgement at former classification bill bundle gray-scale value.
For step 4)
Because the gray level difference of waist bar and strapping and banknote itself at first extracts waist bar and strapping position.According to the condition that the position may occur, the method for taking to travel through is estimated the situation of waist bar, and passes through the traversal back different evaluation of weight as a result, the situation of finally determining to tie up the waist bar.
Described what extract that waist bar and strapping position use is self-adaption binaryzation method, again binary image is carried out the connected domain analysis, has gray difference according to waist bar and strapping and banknote itself, thereby can determine their position by this species diversity.
The condition that the position may occur comprises: most of regional waist bar to single bundle banknote is wrapped in 2/3 position of whole banknote, and the mode that 10 bundle banknote waist bars are above 5 times 5 is divided right and left.Area in addition is wrapped in 1/2 position with the unification of waist bar, breakthrough status occurs behind the 10 bundle banknote bundlings.To discarded banknote, the unified two waist bars of 1/3 position that use are tied up, and whole bundle strapping material difference, exist pure white strapping to run through whole waist bar.
Is number identification to obtain by the shadowing analysis between every handle, through determining the state of number judgement influence that the strapping that weakens simultaneously brings when pushing down the waist bar and waist bar deformation effect with the cross assessment of texture analysis.Under the condition of number deficiency, increase every thickness average value relatively, to avoid because the erroneous judgement that the two bundle banknote waist bars coincidences that the distortion of waist bar produces bring.
And the cross assessment of texture analysis is that the dash area between every handle is carried out projection by horizontal direction, the result who sees projection 9,10 and 11 immediate under three kinds of situations be that is a kind of.
Every thickness average value relatively be see about under each situation of 5, it is consistent that the average thickness on both sides is wanted.
For step 4)
In identifying, use red channel,, change the curve of certain passage separately because for a set image, can cause colour cast, according to actual conditions, the name chapter on the waist bar is red, use red channel, can eliminate the difference that the unknown chapter image in famous Zhanghe brings.By an implicit name chapter, with the shade between image normalization and the outstanding every bundle, promptly a name Zhang Yanse will desalinate and be white in the image red channel, improve card accuracy of identification.
For the situation of brand-new banknote, because brand-new stiff thickness strict conformance will stack in strict accordance with last 5 times 5 cross methods.After in algorithm identified, removing the original analytical approach of use, also need the comparison of connected domain area, to promote the precision of test result.This is because because fresh money can produce the higher pixel of gray-scale value, so under each situation of 5 of both sides, judge whether the connected domain area of both sides gray-scale value upper zone equates.
During concrete Project Realization, can utilize existing programming tool to carry out with reference to Figure 10 flow process.
Claims (9)
1. a machine vision inspection method for paper money is characterized in that reading in image with image-pickup device earlier, and what image frame showed is banknote thickness, and the banknote thickness direction is made as the x direction, and the width of paper money direction is perpendicular to picture, and the length direction of bank note is made as the y direction; Again to the waist bar card of bundled bank note whether correctly detecting, concrete steps comprise:
1) connected domain analysis: choose suitable judgement image earlier, and choose largest connected territory, i.e. target area; Judge tentatively according to largest connected territory whether image is correct, correctly then enter next step;
2) seek flex point in the target area the corner of pack of paper coin, if find four flex points, then enter next step;
3) judge folder: respectively edge by the evaluating objects zone or texture judge whether folder, if having folder, finish to judge that the prompting folder is withdrawing from; If the folder handle does not then enter next step;
4) judge many handles or few handle: obtain white waist bar information in the target area earlier; Judge whether that by analyzing the waist bar number is correct; If mistake, then prompting is miscounting mistake, withdrawing from; If correct, then finish to judge that prompting right withdraws from;
In the described step 3), comprise step:
3.1) edge in evaluating objects zone: for the image that obtains flex point, earlier image is carried out texture analysis, promptly the texture that the thickness of bank note constitutes on the x direction is analyzed; Move towards stack to what analyzing image texture obtained, judge that banknote stacks direction; Then the both sides image edge addition that is judged to be the correspondence that stacks direction; Again the image edge after the stack is analyzed, then determined to exist the folder handle;
3.2) step 3.1) judge that conclusion is not press from both sides handle, then jump out step 3), if folder handle, the then texture of further analysis image are arranged:
When if texture is straight line, step 3.1) folder that obtains the result for false; Texture is a straight line, and promptly image edge in the left and right sides adds up to straight line, do not protrude or recessed, about the phase side uneven owing to tie up out-of-flatness and cause;
If the texture in the x direction is a curve, and bending direction is identical with the direction of judging the folder handle, and promptly texture is bent downwardly and adjudicates when pressing from both sides the position below image, and folder is the result false;
If the texture in the x direction is a curve, and the direction of bending direction and judgement folder handle is opposite, and promptly texture is bent downwardly and adjudicates when pressing from both sides the position above image, and folder is the result true.
2. method according to claim 1 is characterized in that in the described step 1), and it is dark forcing background earlier, by strengthening picture contrast and image binaryzation, is partitioned into image boundary;
Promptly come mode that the texture of image is followed the tracks of by texture tracked again, analyze image zones of different deformation extent, and adjust original image state to the distortion by image target area internal tangent direction; Described this tangential direction is the direction vertical with the gradient direction of image;
Described texture tracked concrete steps are, difference according to the bill bundle deformation texture degree that analyzes, the influence of crushed element is remedied, in general because the strength trend of tying up, the edge part branch on image both sides upwarps or has a downwarp, and length direction can shorten, and according to deformation extent the artificial compensation of shortening of part is got on.
3. method according to claim 2 is characterized in that in the described step 1), and described correct image is to find a connected region that similarly meets the rectangle closure of general pattern size with all bill bundle samples of learning;
Largest connected territory is to use homomorphic filtering and dynamic histogram thresholding to analyze the method for alternate analysis, accurately sets display foreground and background gray level difference below 20; Judge largest connected territory by image binaryzation, and largest connected territory is corroded and expansion process, filter the edge burr; Under the state at the no image of judgement edge, use background value and the maximum edge of waist bar preserved to simulate the sample edge again.
4. method according to claim 3 is characterized in that in the described step 1)
A, homomorphic filtering: piece image f (x y) can regard as by two component combination and forms, promptly f (x, y)=i (x, y) .r (x, y), (x y) for luminance component is the incident component, is the light intensity that incides on the scenery to i; R (x y) is reflecting component, is the light intensity that is subjected to the scenery reflection, and concrete steps are as follows:
A1) taken the logarithm simultaneously in the both sides of following formula earlier, promptly Inf (x, y)=Ini (x, y)+Inr (x, y);
A2) Fourier transform is got on the following formula both sides, F (u, v)=I (u, v)+R (u, v);
A3) (u, (u v), gets v) to handle F with a frequency-domain function H
H(u，v)F(u，v)＝H(u，v)I(u，v)+H(u，v)R(u，v)；
A4) inverse Fourier transform is to spatial domain, Hff (x, y)=hi (x, y)+hr (x, y); As seen getting image after strengthening is to get two parts by corresponding luminance component and reflecting component to be formed by stacking;
A5) again index is got on the formula both sides in the step (4), homomorphic filtering function g (x, y)=exp|hff (x, y) |=exp|hi (x, y) |+exp|hr (x, y) |; , it can act on luminance component and the reflecting component respectively;
B, dynamically histogram thresholding analysis alternate analysis:
That intersection concave point that adopts two mountain peaks to intersect is followed the threshold value of background as distinguishing sample; Because the difference of external environment condition, this concave point of image that different conditions is taken down is the value of a dynamic change, so to every pictures, all will get the threshold value of the point of this dynamic change as judgement sample and background.
5. method according to claim 1 is characterized in that described step 2) in, the method for seeking flex point is that employing is promptly at first found two the longest points of distance on the edge with image 45 degree angle translation search plans, relocates out two remaining points;
For the banknote flange of 4 angles generations or the seal flange of center section generation, adjust the disturbing factor of individual generation of elimination by once corroding and expanding, and determine banknote flange, the seal flange position that the manual waist bar afterbody that twines leans out or center section produces by the image gray levels analysis once more, by direct removing parts of images, eliminate the influence that flange brings.
6. method according to claim 1, it is characterized in that described step 3.1) concrete steps be, addition is carried out to the projected length at image boundary place in banknote both sides of the edge in the image obtain an one-dimension array, again this one-dimension array is analyzed, judge according to its fluctuation range whether its edge has protrusion or recessed, and have only when protrude or recessed continuous part length greater than the both sides of the edge overall length 10% the time, just determine to exist folder.
7. method according to claim 1 is characterized in that in the described step 4),
White waist bar information obtains by self-adaption binaryzation in a, the target area:
Because the gray level difference of waist bar and strapping and banknote itself, at first extract white waist bar and strapping position, specifically be usefulness be that self-adaption binaryzation method obtains, again binary image is carried out the connected domain analysis, there is gray difference according to waist bar and strapping and banknote itself, determines their position by this species diversity;
The condition that may occur again according to the position, the method for taking to travel through is estimated the situation of waist bar, and by the traversal back different evaluation of weight as a result, the situation of finally determining to tie up the waist bar;
B, be number identification what obtain by the shadowing analysis between every handle:
The cross assessment of process and texture analysis is promptly undertaken projection to the dash area between every handle by horizontal direction, the result who judges projection 9,10 or 11 immediate under three kinds of situations be that is a kind of, determine the state of number judgement influence that the strapping that weakens simultaneously brings when pushing down the waist bar and waist bar deformation effect;
Under the condition of number deficiency, increase every thickness average value is more promptly judged about under each situation of 5, it is consistent that the average thickness on both sides is wanted.
8. according to claim 1 or 7 described methods, it is characterized in that in the described step 4), use red channel to reach the purpose of implicit name chapter in identifying, with the shade between image normalization and the outstanding every bundle, concrete grammar is that a name Zhang Yanse will desalinate and be white in the image red channel.
9. method according to claim 8 is characterized in that in the described step 4), for brand-new banknote, stacks in strict accordance with last 5 times 5 cross methods, and because brand-new stiff thickness strict conformance; In algorithm identified, also increased connected domain area comparison process: because fresh money can produce the higher pixel of gray-scale value, so, judge whether the connected domain area of both sides gray-scale value upper zone equates under each situation of 5 of both sides.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010101036445A CN101799868B (en) | 2010-02-01 | 2010-02-01 | Machine vision inspection method for paper money |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010101036445A CN101799868B (en) | 2010-02-01 | 2010-02-01 | Machine vision inspection method for paper money |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101799868A true CN101799868A (en) | 2010-08-11 |
| CN101799868B CN101799868B (en) | 2012-05-30 |
Family
ID=42595540
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010101036445A Active CN101799868B (en) | 2010-02-01 | 2010-02-01 | Machine vision inspection method for paper money |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101799868B (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102298695A (en) * | 2011-08-09 | 2011-12-28 | 深圳捷佳德现金自动化处理设备有限公司 | Visual analyzing and processing method for detecting paper money bundle |
| CN102637317A (en) * | 2012-04-26 | 2012-08-15 | 东南大学 | Coin size measuring method based on vision |
| CN103971159A (en) * | 2014-05-05 | 2014-08-06 | 石家庄印钞有限公司 | Whole bundle of bank note number of passing detection device |
| CN104528063A (en) * | 2014-10-30 | 2015-04-22 | 中钞长城金融设备控股有限公司 | Device and method for detecting flake-shaped materials not folded together |
| CN104943891A (en) * | 2015-06-04 | 2015-09-30 | 浙江大学台州研究院 | Paper currency shaping and bundling device |
| CN105015856A (en) * | 2015-06-04 | 2015-11-04 | 浙江大学台州研究院 | Detection method for paper money shaping bundling |
| CN106023174A (en) * | 2016-05-16 | 2016-10-12 | 南昌印钞有限公司 | Novel device and method for detecting the number of bundles of thousand-banknote packet |
| CN107204069A (en) * | 2016-03-16 | 2017-09-26 | 深圳市创自技术有限公司 | Paper currency identifier, the method for Paper currency identifier identification width of paper money and self-service device machine |
| CN108596192A (en) * | 2018-04-24 | 2018-09-28 | 图麟信息科技（深圳）有限公司 | A kind of face amount statistical method, device and the electronic equipment of coin code heap |
| CN111310628A (en) * | 2020-02-10 | 2020-06-19 | 武汉科技大学 | Paper currency forming mode inspection and identification method based on paper currency printing pattern characteristics |
| CN111879781A (en) * | 2020-06-17 | 2020-11-03 | 西安交通大学 | Test paper quality detection system based on machine vision |
| CN112036465A (en) * | 2020-08-26 | 2020-12-04 | 中国建设银行股份有限公司 | Image recognition method, device, equipment and storage medium |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100583172C (en) * | 2005-11-04 | 2010-01-20 | 中国印钞造币总公司 | Finance bills detection system |
| CN200959137Y (en) * | 2006-09-01 | 2007-10-10 | 沈阳中钞信达金融设备有限公司 | Banknote or bill image fetch |
| CN201081690Y (en) * | 2007-09-07 | 2008-07-02 | 中钞长城金融设备控股有限公司 | Banknote thickness tester |
-
2010
- 2010-02-01 CN CN2010101036445A patent/CN101799868B/en active Active
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102298695B (en) * | 2011-08-09 | 2013-03-20 | 深圳捷佳德现金自动化处理设备有限公司 | Visual analyzing and processing method for detecting paper money bundle |
| CN102298695A (en) * | 2011-08-09 | 2011-12-28 | 深圳捷佳德现金自动化处理设备有限公司 | Visual analyzing and processing method for detecting paper money bundle |
| CN102637317A (en) * | 2012-04-26 | 2012-08-15 | 东南大学 | Coin size measuring method based on vision |
| CN103971159B (en) * | 2014-05-05 | 2017-05-24 | 石家庄印钞有限公司 | Whole bundle of bank note number of passing detection device |
| CN103971159A (en) * | 2014-05-05 | 2014-08-06 | 石家庄印钞有限公司 | Whole bundle of bank note number of passing detection device |
| CN104528063A (en) * | 2014-10-30 | 2015-04-22 | 中钞长城金融设备控股有限公司 | Device and method for detecting flake-shaped materials not folded together |
| CN104528063B (en) * | 2014-10-30 | 2018-06-22 | 中钞长城金融设备控股有限公司 | For detecting the device and method that flaky material dissipates bundle |
| CN104943891B (en) * | 2015-06-04 | 2018-05-04 | 浙江大学台州研究院 | Bank note sizing binding apparatus |
| CN105015856A (en) * | 2015-06-04 | 2015-11-04 | 浙江大学台州研究院 | Detection method for paper money shaping bundling |
| CN104943891A (en) * | 2015-06-04 | 2015-09-30 | 浙江大学台州研究院 | Paper currency shaping and bundling device |
| CN105015856B (en) * | 2015-06-04 | 2018-07-06 | 浙江大学台州研究院 | Detection method is tied up in bank note sizing |
| CN107204069A (en) * | 2016-03-16 | 2017-09-26 | 深圳市创自技术有限公司 | Paper currency identifier, the method for Paper currency identifier identification width of paper money and self-service device machine |
| CN107204069B (en) * | 2016-03-16 | 2020-09-08 | 深圳市创自技术有限公司 | Paper money recognizer, method for recognizing paper money width by paper money recognizer and self-service equipment |
| CN106023174A (en) * | 2016-05-16 | 2016-10-12 | 南昌印钞有限公司 | Novel device and method for detecting the number of bundles of thousand-banknote packet |
| CN106023174B (en) * | 2016-05-16 | 2019-09-13 | 南昌印钞有限公司 | A kind of novel detection money thousand sheets packet is several devices and methods therefors |
| CN108596192A (en) * | 2018-04-24 | 2018-09-28 | 图麟信息科技（深圳）有限公司 | A kind of face amount statistical method, device and the electronic equipment of coin code heap |
| CN111310628A (en) * | 2020-02-10 | 2020-06-19 | 武汉科技大学 | Paper currency forming mode inspection and identification method based on paper currency printing pattern characteristics |
| CN111879781A (en) * | 2020-06-17 | 2020-11-03 | 西安交通大学 | Test paper quality detection system based on machine vision |
| CN111879781B (en) * | 2020-06-17 | 2022-02-11 | 西安交通大学 | Test paper quality detection system based on machine vision |
| CN112036465A (en) * | 2020-08-26 | 2020-12-04 | 中国建设银行股份有限公司 | Image recognition method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101799868B (en) | 2012-05-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101799868B (en) | Machine vision inspection method for paper money | |
| Kumar et al. | A fast DCT based method for copy move forgery detection | |
| EP3089103B1 (en) | Method for correcting fragmentary or deformed quadrangular image | |
| US20140270460A1 (en) | Paper identifying method and related device | |
| Ren et al. | Fusion of intensity and inter-component chromatic difference for effective and robust colour edge detection | |
| CN104077594B (en) | A kind of image-recognizing method and device | |
| Qu et al. | Detect digital image splicing with visual cues | |
| CN102663357A (en) | Color characteristic-based detection algorithm for stall at parking lot | |
| DE10123330A1 (en) | Detection of falsified fingerprints, e.g. a silicon casting of a fingerprint, using a dynamic, software-based method for detection of falsified fingerprints that is quick and efficient | |
| CN103455820A (en) | Method and system for detecting and tracking vehicle based on machine vision technology | |
| CN104298989A (en) | Counterfeit identifying method and counterfeit identifying system based on zebra crossing infrared image characteristics | |
| Zach et al. | Automated image forgery detection through classification of JPEG ghosts | |
| CN109344886B (en) | Occlusion number plate distinguishing method based on convolutional neural network | |
| RU2745098C2 (en) | Method of determinig authorship of a painting | |
| JP7226493B2 (en) | Contact wire wear inspection method | |
| CN101303728A (en) | Method for identifying fingerprint facing image quality | |
| Geusebroek et al. | Learning banknote fitness for sorting | |
| CN104408728A (en) | Method for detecting forged images based on noise estimation | |
| CN104568956A (en) | Machine vision based detection method for strip steel surface defects | |
| CN106599923B (en) | Method and device for detecting seal anti-counterfeiting features | |
| CN115620075A (en) | Method, system and equipment for generating data set for leukocyte classification model | |
| CN103927512B (en) | vehicle identification method | |
| Julliand et al. | Automated image splicing detection from noise estimation in raw images | |
| CN113205633B (en) | Method and device for detecting red and blue fibers of bill, electronic equipment and storage medium | |
| CN102592151B (en) | Blind detection method for median filter in digital image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |