CN100385450C - Vehicle license plate extraction method based on high-hat switch and wavelet switch - Google Patents
Vehicle license plate extraction method based on high-hat switch and wavelet switch Download PDFInfo
- Publication number
- CN100385450C CN100385450C CNB2005100215409A CN200510021540A CN100385450C CN 100385450 C CN100385450 C CN 100385450C CN B2005100215409 A CNB2005100215409 A CN B2005100215409A CN 200510021540 A CN200510021540 A CN 200510021540A CN 100385450 C CN100385450 C CN 100385450C
- Authority
- CN
- China
- Prior art keywords
- value
- horizontal
- order difference
- row
- car plate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Abstract
The invention relates to a license plate positioning method based on cap transformation and wavelet transformation, wherein it uses horizontal positioning and vertical positioning, that using cap transformation to protrude the license plate area, to horizontally position the license plate; then using wavelet transformation to extract the character of license plate, to realize the vertically position of license plate; at last, based on obtained license plate horizontal position and vertical position, extracting the license plate from original image. The invention can extract out the high-frequency information of license plate most, to accurately position the license plate, to improve its environment adaptability, and it can improve the calculation speed.
Description
Technical field
The invention belongs to the image processing technique field, particularly the vehicle license plate extraction method in the complex background in the license plate recognition technology.
Background technology
Intelligent transportation is the main direction of current traffic administration development, is the forward position research topic of present world traffic and transport field.The license plate automatic identification technology then is the core of intelligent transportation system.It is the important means that solves the freeway management problem, is the application at intelligent transportation field of computer image processing technology and mode identification technology.This technology is in the problems that solve highway, and as vehicle toll and management, the magnitude of traffic flow detects, parking lot fee collection management, and monitoring vehicle breaking regulation is widely used in the particular problems such as fake license vehicle identification, has huge economic and realistic meaning.Simultaneously, it occupies critical role in project managements such as urban road, harbour and airport.Along with the raising of computing power and the development of image processing techniques, Vehicle License Plate Recognition System reaches its maturity.See document for details: T.Vaito, T.Tsukada, K.Yamada, K.Kozuka, andS.Yamamoto, " Robust license-plate recognition method for passing vehicles under outsideenvironment, " IEEE Trans.Veh.Technol., vol.49, pp.2309-2319, Nov.2000 and document: Shyang-Lih Chang, Li-Shien Chen, YunChung Chung, Sei-Wan Chen, Automatic license platerecognition, Intelligent Transportation Systems, IEEE Transactions on, March 2004 is described.
In automatic license plate recognition technology, the location of car plate is the emphasis and the difficult point of whole recognition technology.The positioning time of car plate and precision directly influence the performance of whole Vehicle License Plate Recognition System.In real life, owing to be subjected to the complicacy of background, the unevenness of illumination condition and the environmental factors such as ambiguity of weather conversion, and the influence of factor such as the inclination program of car plate own, contaminated degree, present most license plate locating methods just are confined to some side or limit under given conditions, just can finish the accurate location to car plate.General, can be applicable to that the license plate locating method of all environment and condition does not also successfully find out.Therefore, how to have now on all valuable achievements in research, improving versatility, the shortening positioning time of car plate positioning system and improve the main direction that bearing accuracy will become our current research.See document for details: Chacon M, M.I.Zimmerman.A License plate location based on a dynamic PCNN scheme Neural Networks.Proceedings of the International Joint Conference, 20-24 July 2003,1195-1200 vol.2 and document: D.Irecki ﹠amp; D.G.Bailey, " Vehicle registration plate localization and recognition ", Proceedings ofthe Electronics New Zealand Conference, ENZCon ' 01, New Plymouth, New Zealand, September2001
The method that present normally used car plate extracts has:
(1) based on the vehicle license plate extraction method of scan line.It through the characteristic that license plate area can clocklike rise and fall, reaches the purpose of identification license plate area by scan line.Its shortcoming is to be applicable to that license plate image brightness changes situation relatively uniformly.But license plate image is in the environment of variable light source in actual applications, and brightness is extremely inhomogeneous, thereby has limited the use of said method.See document Agui T for details, Choi H J, Kajima N.Method of extracting car number plates by imageprocessing[J] .System and Computers, 1998,19 (3): 45～52
(2) based on the vehicle license plate extraction method of colour.It is to distinguish license plate area and background area by other regional rainbow features that are different from of extracting car plate, thereby extracts car plate.Its shortcoming is to be difficult to solve locating accuracy problem under the illumination unevenness environment, and locating speed is slow, is difficult to reach the requirement of real-time identification.See document Davies P for details, EmmottN, Ayland N.License plate recognition technology for toll violation enforcement[J] .Proceedingsof IEEE Colloquium on Image Analysis for Transport Applications, 1990,35 (2): 711～715
(3) based on the method for rim detection.It is analyzed by extracting image edge information, and then records the car plate edge.Its shortcoming is that the continuity of requirement image border will be got well, but the car plate frame of actual photographed is often discontinuous.See document Fu Yuqing for details, Shen Wei, Huang Xinhua.Research on vehicle license plate character extractionfrom complex background[J] .Pattern Recognition and Artificial Intelligence, 2000,13 (3): 345～348 (in Chinese)
(4) based on Artificial Neural Network model.It utilizes self-adaptation, the self-learning capability of neural network, reaches the purpose of car plate identification by training.Its advantage is to melt some pre-service and discern in one, recognition speed fast; Shortcoming is when characterizing definition is met difficulty, and effect can not be satisfactory.See document Rausm for details, Kreftl.Reading car license plates bythe use of artificial neural networks In:Proceedings of the 1995 IEEE 38
ThMidwest Symposiumon Circuits and Systems, NJ, USA:IEEE 1995, Part 1 (of2): 538-541.
The common ground of four kinds of above-mentioned vehicle license plate extraction methods is: these methods all are at a certain conditions, are subjected to the restriction of factors such as weather, background, illumination easily, and robustness is bad.In case condition changes, they cut apart accuracy rate bigger fluctuation will take place, thus the performance of whole Vehicle License Plate Recognition System reduces greatly.
Summary of the invention
Task of the present invention provides a kind of vehicle license plate extraction method based on cap transformation and wavelet transformation, and it has characteristics such as bearing accuracy height and highly versatile.
In order to describe content of the present invention easily, at first introduce several notions, and some terms are defined.
Notion one: mathematical morphology.Mathematical morphology is based on graphical analysis, and the form of removing the dimensioned plan picture with " structural element " with certain morphosis is to solve the image understanding problem.Morphologic basis is corrosion and dilation operation, and the open and close computing that produces therefrom.The formula of corrosion and dilation operation is respectively:
The formula of open and close computing is respectively:
Wherein, A is an original image, and B is a structural element, and U is the image that original image obtains after corrosion, and V is the image that original image obtains after expanding, and Θ is the erosion operation symbol,
Be the dilation operation symbol, ο is the opening operation symbol, is the closed operation symbol, and Φ is the empty set symbol.
Notion two: high cap (Top-hat) conversion.Cap transformation is a kind of efficient mathematical morphological transformation.Its principle is exactly to deduct it is made the image that obtains behind the opening operation from a width of cloth original image, and transformation for mula is: HAT (A)=A-(A ο B).Wherein A is an original image, and B is a structural element, opening operation
HAT (A) is the image of original image through obtaining behind the cap transformation.This transfer pair is asked dark set of pixels or is asked bright set of pixels very effective in darker background in brighter background, and situation such as, inclination damaged to noise, illumination conversion, licence plate and distortion is insensitive, can realize quick and precisely cutting apart of licence plate.
Notion three: wavelet transformation.Wavelet transformation is a kind of time-yardstick (T/F) analytical approach of signal.Its basic ideas are by flexible and translation original signal to be decomposed into a series of different spatial resolutions that have, the subband signal of different frequency characteristic and directivity characteristics.These subband signals have characteristics such as good time domain and frequency domain, and these characteristics can be used for representing the local feature of original signal.
Definition 1: car plate.Hang on the front end of vehicle or the number plate that is used to identify testing vehicle register at rear portion.Its outward appearance is a rectangle, and unified dimensions is arranged, and comprises 7 characters altogether.The vehicle of different purposes, the standard of car plate may be different.
Definition 2:R, G, B component.R, G, B component are meant three kinds of primary colours of red, green, blue of forming a width of cloth coloured image.Wherein R represents redness, and G represents green, and B represents blue.
Definition 3: gray level image.Only comprise monochrome information in the image and without any the image of other colouring informations.
Definition 4: greyscale transformation.A kind of coloured image is converted into the mapping mode of gray level image, its transformation for mula is: and f (i, j)=0.114*I (i, j, 1)+0.587*I (i, j, 2)+0.299*I (i, j, 3).Wherein, the line position of i presentation video; The column position of j presentation video; F (i, j) gray-scale value of the pixel of the capable j row of i in the gray level image after the expression conversion; I (i, j, 1), I (i, j, 2) and I (i, j, 3) represent the R of the pixel of the capable j row of i in the coloured image, G, the value of B component respectively; * be multiplication sign.
Definition 5: horizontal first order difference.In the image, the gray-scale value of a back pixel of each row deducts the gray-scale value of previous pixel, obtains the horizontal first order difference value of this image.Its formula is: g
v(i, j)=| f (i, j+1)-f (i, j) |.Wherein, the line position of i presentation video, the column position of j presentation video, f (i, the j) gray-scale value of the pixel of the capable j row of expression i, f (i, the j+1) gray-scale value of the pixel of the capable j+1 row of expression i, g
v(i, j) expression corresponding horizontal first order difference.Horizontal first order difference can be given prominence to the vertical detail information of image, is convenient to edge extracting.
Definition 6: the horizontal projection of the horizontal difference of single order.A kind ofly add up by horizontal direction, reality is transformed into method in the one-dimensional space to the gradation of image level error score value in the two-dimensional space, and this transforming function transformation function is
。G wherein
v(i, j) the horizontal first order difference value of the capable j row of expression i, T
H(i) be the capable horizontal projection value of i, the variation range of j is from 1 to n, the length of n representative image.
Definition 7: Gaussian filter.One of the most frequently used smoothing filter in the Flame Image Process is made convolution by Gaussian function and first order difference horizontal projection and is come burr in the level and smooth perspective view, reaches filtering and level and smooth effect.Filter function is
T ' wherein
H(i) be filtered first order difference horizontal projection value, the variation range of i is from 1 to n, the length of n representative image; W represents the width size of smooth region, gets 8 herein;
Be Gaussian function,
σ represents the mean square deviation of gray level image; T
H(i) i first order difference horizontal projection value of expression, T
H(i-j) expression (i-j) individual first order difference horizontal projection value, T
H(i+j) expression (i+j) individual first order difference horizontal projection value.
Definition 8: binaryzation process.The all values of entire image is changed into the process of having only two kinds of values, and generally these two kinds of values are 0 and 1 or 0 and 255.Its concrete method is: when the value on the image was greater than or equal to the binaryzation threshold values, the value of this point was turned to 1 (or 255) by two-value; When the value on the image less than the binaryzation threshold values time, the value of this point is turned to 0 by two-value.
Definition 9: the threshold values of binaryzation.Selected thresholding when image is carried out binaryzation.The computing formula of binaryzation threshold values is: T=t*aver, and wherein T is the binaryzation threshold values, and aver is the average through the first order difference horizontal projection behind the gaussian filtering, and t is weights, and * is a multiplication sign.Select for use suitable binaryzation threshold values can realize the purpose of outstanding license plate area, removal noise.
Definition 10: car plate horizontal location.Determine the horizontal level of license plate area in entire image, promptly determine the coboundary of car plate and the process of lower boundary.Because the background color of car plate and the color of car plate word form sharp contrast, and frequent in relatively little scope inner conversion, so the first order difference horizontal projection value of car plate horizontal level is bigger.
Definition 11: car plate horizontal level candidate region.The one or more strip regions that may comprise car plate that after the car plate horizontal location, obtain.These zones have bigger first order difference horizontal projection value, and peak width is limited between the certain limit, approximate the width of car plate, the coboundary in zone is the coboundary of car plate horizontal level candidate region, and the lower boundary in zone is the lower boundary of car plate horizontal level candidate region.
Definition 12: fast wavelet transform.Fast wavelet transform (Mallat) is a kind of efficient calculation that realizes wavelet transform (DWT), and the relation of the practicality between adjacent yardstick DWT coefficient has been found in this conversion.Concrete conversion principle is seen accompanying drawing 1.
Definition 13: car plate is vertically located.Determine the vertical position of license plate area in entire image, i.e. the process of the left margin of car plate and right margin.
A kind of license plate locating method provided by the invention based on cap transformation and wavelet transformation, it comprises the following step:
The camera head of the appropriate location of step 1. by being installed on highway crossing or parking lot carries out image acquisition to the vehicle that enters in the image pickup scope, obtains containing the original image of license plate image.
Horizontal first order difference figure behind the step 5. pair cap transformation carries out horizontal projection, obtains the horizontal projection curve of first order difference.The computing formula of the horizontal projection of first order difference is
G ' wherein
v(i, j) expression is through the horizontal first order difference value of the capable j row of the i behind the cap transformation, T
H(i) be the capable horizontal projection value of i, the variation range of j is from 1 to n, the length of n representative image.
First order difference horizontal projection curve in the step 6. pair step 5 carries out gaussian filtering, obtains a level and smooth first order difference horizontal projection curve.In practical methods, adopt discrete Gauss's smoothing method that projection value is carried out smoothing processing, concrete grammar is as follows
T ' wherein
H(i) be filtered first order difference horizontal projection value; The variation range of i is to n from 1; The length of n representative image; W represents the width size of smooth region, gets 8 herein;
It is Gaussian function;
σ represents the mean square deviation of gray level image; T
H(i) i first order difference horizontal projection value of expression; T
H(i-j) expression (i-j) individual first order difference horizontal projection value; T
H(i+j) expression (i+j) individual first order difference horizontal projection value.
Step 7. is calculated the threshold values of binaryzation.The computing formula of binaryzation threshold values is: T=t*aver.Wherein T is the threshold values of binaryzation; Aver is the average through the first order difference horizontal projection behind the gaussian filtering; T is weights, span between 0.8～1.2, binaryzation best results when the value of t is 1.1.
The threshold values that step 8. utilizes step 7 to calculate carries out binaryzation to passing through the first order difference horizontal projection curve that obtains behind the gaussian filtering in the step 6, obtains the binary picture of first order difference horizontal projection curve.Binarization method is: more than or equal to threshold values, just the value at this place is set to 1 as if the value on the first order difference horizontal projection curve, otherwise the value at this place is set to 0.Through binaryzation, first order difference horizontal projection curve is converted into the sequence of forming by a series of 0 and 1.Represent the bigger zone of first order difference horizontal projection value by the 1 continuous sequence of forming; Represent the less zone of first order difference horizontal projection value by the 0 continuous sequence of forming; The border in two kinds of zones is represented in 0 to 1 or 1 to 0 saltus step, respectively 0 to 1 and 1 to 0 saltus step location records in storehouse stack1 and stack2.
Step 9. utilizes binary picture to carry out the car plate horizontal location, determines car plate horizontal level candidate region.1 continuous in the binary picture sequence of forming is car plate horizontal level candidate region; Value among the stack1 is the left margin of car plate horizontal level candidate region, i.e. the coboundary top of car plate; Value among the stack2 is the right margin of car plate horizontal level candidate region, i.e. the lower boundary bottom of car plate; The value of length=stack2-stack1 is the width of car plate horizontal level candidate region.The value of rejecting length is greater than 100 with less than 6 pseudo-car plate horizontal level candidate region, and remaining zone is final car plate horizontal level candidate region.
Step 10. utilizes the quick wavelet-decomposing method of Mallat that wavelet decomposition is carried out in car plate horizontal level candidate region, obtains four number of sub images B, H, V and the D of car plate horizontal level candidate region.The quick wavelet decomposition flow process of Mallat is: A makes wavelet transformation to input picture, obtains 4 subgraphs, is defined as level and smooth subgraph B, horizontal subgraph H, vertical subgraph V and oblique subgraph D respectively; Wherein, horizontal subgraph H along continuous straight runs allows low frequency component to pass through, and vertically allows high fdrequency component to pass through; Vertical subgraph V along continuous straight runs allows high fdrequency component to pass through, and vertically allows low frequency component to pass through; Oblique subgraph D along continuous straight runs and vertical direction all allow high fdrequency component to pass through.
The method that step 11. utilizes step 3 to provide, the horizontal first order difference of calculating H, V and D three number of sub images obtains three horizontal first order difference images.
Step 12. obtains the horizontal first order difference summation of Wavelet image of car plate to the horizontal first order difference respective items addition respectively of H, V and D three number of sub images.Concrete grammar is for adopting formula Dif_I (i, j)=| H (i, j+1)-H (i, j) |+| V (i, j+1)-V (i, j) |+| D (i, j+1)-D (i, j) | carry out the corresponding sum operation of horizontal first order difference, wherein H (i, j) the horizontal first order difference value of the capable j row of the i of expression H subgraph, H (i, j+1) the horizontal first order difference value of the capable j+1 row of the i of expression H subgraph, V (i, j) the horizontal first order difference value of the capable j row of the i of expression V subgraph, V (i, j+1) the horizontal first order difference value of the capable j+1 row of the i of expression V subgraph, D (i, j) the horizontal first order difference value of the capable j row of the i of expression D subgraph, D (i, j+1) the horizontal first order difference value of the capable j+1 row of the i of expression D subgraph, Dif_I (i, j) the small echo subgraph first order difference summation of the car plate of the capable j row of expression i.
Step 13. is calculated the global density and the row density of small echo subgraph first order difference summation.Global density is meant the number of nonzero value among the matrix D if_I, its computing method are: counter counter is set, from top to bottom, scan matrix Dif_I from left to right, when running into nonzero value, the value of counter is added 1, and the value among the counter that obtains during been scanned is exactly the value of global density.Row density is meant the non-zero number of each row among the matrix D if_I.The total n row of matrix D if_I, n is the height of image.The computing method of row density are: be provided with n counter counter_1, counter_2 ..., counter_n, counter_1 is used for the nonzero value number of count matrix Dif_I first row, counter_2 is used for the nonzero value number of count matrix Dif_I secondary series, the rest may be inferred, counter_n is used for the nonzero value number of count matrix Dif_I n row, each row among the scan matrix Dif_I from top to bottom, when running into nonzero value, the value of the counter of respective column is added 1, and the value in the individual count device that obtains during been scanned is exactly the row density of each row.At last, the row density matrix pos that 1*n dimension is set preserves the value of row density.
Step 14. is determined the threshold values of row density binaryzation, and binaryzation row density.With 0.2 times global density is that threshold values carries out binaryzation to row density matrix pos, and the method for binaryzation is: the value more than or equal to threshold values among the row density matrix pos is turned to 1 by two-value, is turned to 0 less than the value of threshold values by two-value.Thus, row density matrix pos is the sequence of forming by a series of 0 and 1 by binaryzation.Represent the zone that row density is bigger by the 1 continuous sequence of forming; Represent the zone that row density is less by the 0 continuous sequence of forming; The border in two kinds of zones is represented in 0 to 1 or 1 to 0 saltus step, respectively 0 to 1 and 1 to 0 saltus step location records in storehouse stack3 and stack4.
Step 15. is carried out car plate and is vertically located, and determines the left margin and the right margin of car plate.By the binaryzation of step 14, the continuous pairing position of 1 sequence is license plate area column position in entire image among the row density matrix pos, and the value among the stack3 is the left margin left of license plate area, and the value among the stack4 is the right margin right of license plate area.
Step 16. is determined real license plate area according to the characteristic of car plate.Concrete method is: reject the car plate width less than 30 and the car plate width less than the pseudo-license plate area of car plate height, thereby obtain real license plate area.
Step 17. is utilized four boundary values of the license plate area that step 9 and step 15 obtain, and extracts license plate area.Concrete car plate extraction step is as follows: at first according to the value of car plate coboundary top that obtains in the step 9 and lower boundary bottom, calculate the width plate_width=bottom-top of car plate; Car plate left margin left that obtains according to step 15 and the value of right margin right are again calculated the length plate_length=right-left of car plate; Then, size of definition matrix car_plate that is plate_width*plate_length is used for storing and cuts apart later car plate; At last, give car_plate the capable numerical value assignment capable to bottom, that left is listed as the right row of top in the original image, at this moment, matrix car_plate is automotive license plate.
By above step, we just extract license plate image from the original image that contains car plate.
Need to prove:
1., then no longer do gradation conversion and handle if the original vehicle image that includes car plate that collects in the step 2 is a gray level image.
2. carrying out horizontal first order difference calculating in the step 3 is because the pixel value in the license plate area changes soon and be concentrated, and more obviously concentrates on horizontal first order difference aspect.
3. the cap transformation in the step 4 is to asking dark set of pixels or ask bright set of pixels very effective in darker background in brighter background, and car plate just in time has similar character in entire image.Thereby image is carried out cap transformation can give prominence to license plate area.
4. in the step 6,, be unfavorable for the accurate location of method, must adopt Gaussian function that projection value is carried out smoothing processing because the first order difference horizontal projection curve burr that is obtained by step 5 is a lot.
5. the calculating of binaryzation threshold values is for the ease of the first order difference horizontal projection curve behind the gaussian filtering being carried out binaryzation, realizing the purpose of outstanding license plate area, removal noise in the step 7.
6. in the step 9, because the color of the background color of car plate and car plate word forms sharp contrast, and also frequent in relatively little scope inner conversion, so the first order difference horizontal projection value of license plate area is bigger.Therefore, the 1 continuous sequence of forming is car plate horizontal level candidate region in binary picture.
7. in the wavelet transformation of step 10,2-d wavelet function and scaling function can obtain by one dimension wavelet function and the conversion of scaling function process tensor product, picture signal is carried out 2-d wavelet decompose, and its basic decomposition step as shown in Figure 1.Input picture A obtains 4 subgraphs later on through wavelet transformation in Fig. 1, is respectively level and smooth subgraph B, horizontal subgraph H, vertical subgraph V and oblique subgraph D.Wherein, horizontal subgraph H is through twice filtering, and along continuous straight runs allows low frequency component to pass through, and vertically allows high fdrequency component to pass through.This is to strengthen to horizontal line segment, and is level and smooth to vertical line segment.The pixel of vertical direction and tilted direction is highlighted expression respectively by same reason in subgraph V and D.And among the figure,
With
Expression low pass, Hi-pass filter;
The sampling of expression row keeps all even columns;
The expression line sampling keeps all even number lines.
8. the respective items addition in the step 12 is meant that the sum that obtains is kept on the corresponding coordinate position of horizontal first order difference summation matrix row-coordinate identical in the horizontal first order difference matrix of H, V and D three number of sub images and the value addition on the row coordinate.
9. in the step 15, because license plate area has bigger row density, so the continuous pairing position of 1 sequence is license plate area column position in entire image among the row density matrix pos after the binaryzation.
10. in the step 16,, vertically can produce a plurality of license plate candidate areas behind the location through car plate owing in the car plate horizontal location, may produce a plurality of car plate horizontal levels candidate region.Because the length of car plate, width and length breadth ratio restriction are within the specific limits,, obtain real license plate area so can remove pseudo-license plate candidate area according to these characteristics of car plate.
The method that combines based on cap transformation and wavelet transformation that invention proposes can improve performances such as the versatility of system and bearing accuracy effectively.High cap (Top-hat) conversion is exactly to deduct it is made the image that obtains behind the opening operation from a width of cloth original image, this method is to asking dark set of pixels or ask bright set of pixels very effective in darker background in brighter background, and car plate just in time has similar character in entire image.Thereby image is carried out cap transformation can give prominence to license plate area.Wavelet transformation is a kind of time-yardstick (T/F) analytical approach of signal.Because the high-frequency information of license plate area in entire image is obvious and concentrated, thereby wavelet analysis can effectively extract and outstanding these high-frequency informations extract license plate image accurately.
Method of the present invention adopts horizontal location and vertically locatees and combines, and at first adopts cap transformation to give prominence to license plate area, has realized the horizontal location of car plate; By wavelet transformation license plate area is carried out feature extraction then, realized the vertical location of car plate; Last according to the car plate horizontal level and the vertical position information that obtain, from original image, extract car plate.Adopt vehicle license plate extraction method of the present invention, the high-frequency information of extraction license plate area that not only can be as much as possible, thus accurately locate car plate, the environmental adaptation degree of method is improved greatly; And in conjunction with horizontal location and vertically the method for location also improved the computing velocity of method, satisfied the needs of real-time.
Innovation part of the present invention is:
1. combining form is learned relevant knowledge, utilizes cap transformation, and binaryzation is carried out by the image of choose reasonable binaryzation threshold values after to conversion in the position of outstanding license plate area in original image, realizes that the car plate level accurately locatees.
2. utilize wavelet transformation to extract the high-frequency information of candidate region, and, find out license plate image accurately by the Special matrix that scans these high-frequency informations formations then by outstanding these high-frequency informations of horizontal first order difference summation.
3. based on car plate horizontal location and the vertical method that combines of locating,, satisfy the needs of Vehicle License Plate Recognition System real-time so that improve the speed that car plate extracts.At first utilize the horizontal projection of the horizontal first order difference figure behind the cap transformation to carry out the car plate horizontal location, extract a few license plate candidate area, and then these candidate regions are carried out wavelet transformation, the high-frequency characteristic of outstanding license plate image, and then utilize the high-frequency characteristic bianry image of gained to carry out the vertical location of car plate.
Description of drawings
Fig. 1 is the process flow diagram that fast wavelet transform decomposes;
Wherein, A represents the input picture of wavelet transformation, and input is car plate horizontal level candidate region herein; B, H, V and D represent 4 subgraphs behind the wavelet transformation, and B is level and smooth subgraph, and H is horizontal subgraph, and V is vertical subgraph, and D is oblique subgraph.Wherein, horizontal subgraph H is through twice filtering, and along continuous straight runs allows low frequency component to pass through, and vertically allows high fdrequency component to pass through.This is to strengthen to horizontal line segment, and is level and smooth to vertical line segment.The pixel of vertical direction and tilted direction is highlighted expression respectively by same reason in subgraph V and D.
With
Expression low pass, Hi-pass filter;
The sampling of expression row keeps all even columns;
The expression line sampling keeps all even number lines.
Fig. 2 is the original image synoptic diagram that contains car plate;
Wherein, the windshield of 1 expression vehicle; The bonnet of 2 expression vehicles; The zone that 3 expressions are installed car light, hung car plate and bumper bar; 4 expression car plates; 5 expression wheels.In 4 represented car plates, X1, X2, X3, X4, X5, X6 and X7 represent respectively car plate first, second, the 3rd, the 4th, the 5th, the 6th and the 7th character, wherein the spacing of second character and the 3rd character is slightly larger.
Fig. 3 is the license plate image synoptic diagram that the present invention finally obtains;
Wherein, X1, X2, X3, X4, X5, X6 and X7 represent respectively car plate first, second, the 3rd, the 4th, the 5th, the 6th and the 7th character.
Fig. 4 is the process flow diagram of the inventive method.
Specific implementation method:
Adopt method of the present invention, at first use Matlab language compilation car plate identification software, then in the porch of highway, charge station and other any correct positions adopt the original image of camera head automatic shooting vehicle; Then the vehicle original image that photographs is input in the car plate identification software as source data and handles; Obtain the license plate image that a width of cloth comprises recognizable character through the car plate horizontal location with after vertically locating.Adopt 360 that take on the spot, comprise that colored vehicle image under the different conditions such as different weather such as rainy day, greasy weather, fine day and car plate level, license plate sloped, vehicle movement, stationary vehicle is as source data, horizontal location goes out 356, the horizontal location accuracy rate is 98.89%, vertically orient 355, locating accuracy is 98.61%, locatees the license plate image that a width of cloth includes recognizable character and only needs 50ms.
In sum, method of the present invention makes full use of the characteristic of cap transformation and wavelet transformation, in conjunction with the textural characteristics of license plate area, thereby realizes orienting the license plate image that includes recognizable character rapidly and accurately from the vehicle original image that is provided.
Claims (1)
1. license plate locating method based on cap transformation and wavelet transformation is characterized in that it comprises the following step:
The camera head of the appropriate location of step 1. by being installed on highway crossing or parking lot carries out image acquisition to the vehicle that enters in the image pickup scope, obtains containing the original image of license plate image;
Step 2. adopts the greyscale transformation formula that original image is carried out gradation conversion, obtains the gray level image that a width of cloth comprises car plate; Gradation conversion Gongwei f (i, j)=0.114*I (i, j, 1)+0.587*I (i, j, 2)+0.299*I (i, j, 3), the line position of i presentation video wherein, the column position of j presentation video, f (i, j) gray-scale value of the pixel of the capable j row of i in the gray level image after the expression conversion, * is the multiplying symbol, I (i, j, 1), I (i, j, 2) and I (i, j, 3) represent R, the G of the pixel of the capable j row of i in the coloured image, the value of B component respectively;
Step 3. is calculated the horizontal first order difference of gray level image, obtains a horizontal first order difference figure who includes the vehicle gray level image of car plate; Concrete grammar is for adopting formula g
V(i, j)=| f (i, j+1)-f (i, j) | calculate the line position of i presentation video wherein, the column position of j presentation video, f (i, the j) gray-scale value of the pixel of the capable j row of expression i, f (i, the j+1) gray-scale value of the pixel of the capable j+1 row of expression i, g
V(i, j) the horizontal first order difference value of the capable j row of expression i;
Step 4. couple horizontal first order difference figure carries out cap transformation, obtains one through the horizontal first order difference figure behind the cap transformation; The formula of cap transformation is HAT (A)=A-(A о B); Wherein A is horizontal first order difference matrix g
VB is a structural element, and its value all is 1 3*3 matrix for each element; Opening operation
HAT (A) expression is through the horizontal first order difference value g behind the cap transformation
V';
Horizontal first order difference figure behind the step 5. pair cap transformation carries out horizontal projection, obtains the horizontal projection curve of first order difference; The computing formula of the horizontal projection of first order difference is
G wherein
v' (i, j) expression is through the horizontal first order difference value of the capable j row of the i behind the cap transformation, T
H(i) be the capable horizontal projection value of i, the variation range of j is from 1 to n, the length of n representative image;
First order difference horizontal projection curve in the step 6. pair step 5 carries out gaussian filtering, obtains a level and smooth first order difference horizontal projection curve; In practical methods, adopt discrete Gauss's smoothing method that projection value is carried out smoothing processing, concrete grammar is as follows
T wherein
H' (i) be filtered first order difference horizontal projection value; The variation range of i is to n from 1; The length of n representative image; W represents the width size of smooth region, gets 8 herein;
It is Gaussian function;
σ represents the mean square deviation of gray level image; T
H(i) i first order difference horizontal projection value of expression; T
H(i-j) expression (i-j) individual first order difference horizontal projection value; T
H(i+j) expression (i+j) individual first order difference horizontal projection value;
Step 7. is calculated the threshold values of binaryzation; The computing formula of binaryzation threshold values is: T=t*aver; Wherein T is the threshold values of binaryzation; Aver is the average through the first order difference horizontal projection behind the gaussian filtering; T is weights, span between 0.8～1.2, binaryzation best results when the value of t is 1.1;
The threshold values that step 8. utilizes step 7 to calculate carries out binaryzation to passing through the first order difference horizontal projection curve that obtains behind the gaussian filtering in the step 6, obtains the binary picture of first order difference horizontal projection curve; Binarization method is: more than or equal to threshold values, just the value at this place is set to 1 as if the value on the first order difference horizontal projection curve, otherwise the value at this place is set to 0; Through binaryzation, first order difference horizontal projection curve is converted into the sequence of forming by a series of 0 and 1; Represent the bigger zone of first order difference horizontal projection value by the 1 continuous sequence of forming; Represent the less zone of first order difference horizontal projection value by the 0 continuous sequence of forming; The border in two kinds of zones is represented in 0 to 1 or 1 to 0 saltus step, respectively 0 to 1 and 1 to 0 saltus step location records in storehouse stack1 and stack2;
Step 9. utilizes binary picture to carry out the car plate horizontal location, determines car plate horizontal level candidate region; 1 continuous in the binary picture sequence of forming is car plate horizontal level candidate region; Value among the stack1 is the left margin of car plate horizontal level candidate region, i.e. the coboundary top of car plate; Value among the stack2 is the right margin of car plate horizontal level candidate region, i.e. the lower boundary bottom of car plate; The value of length=stack2-stack1 is the width of car plate horizontal level candidate region; The value of rejecting length is greater than 100 with less than 6 pseudo-car plate horizontal level candidate region, and remaining zone is final car plate horizontal level candidate region;
Step 10. utilizes the quick wavelet-decomposing method of Mallat that wavelet decomposition is carried out in car plate horizontal level candidate region, obtains four number of sub images B, H, V and the D of car plate horizontal level candidate region; The quick wavelet decomposition flow process of Mallat is: A makes wavelet transformation to input picture, obtains 4 subgraphs, is defined as level and smooth subgraph B, horizontal subgraph H, vertical subgraph V and oblique subgraph D respectively; Wherein, horizontal subgraph H along continuous straight runs allows low frequency component to pass through, and vertically allows high fdrequency component to pass through; Vertical subgraph V along continuous straight runs allows high fdrequency component to pass through, and vertically allows low frequency component to pass through; Oblique subgraph D along continuous straight runs and vertical direction all allow high fdrequency component to pass through;
The method that step 11. utilizes step 3 to provide, the horizontal first order difference of calculating H, V and D three number of sub images obtains three horizontal first order difference images;
Step 12. obtains the horizontal first order difference summation of Wavelet image of car plate to the horizontal first order difference respective items addition respectively of H, V and D three number of sub images; Concrete grammar is for adopting formula Dif_I (i, j)=| H (i, j+1)-H (i, j) |+| V (i, j+1)-V (i, j) |+| D (i, j+1)-D (i, j) | carry out the corresponding sum operation of horizontal first order difference, wherein H (i, j) the horizontal first order difference value of the capable j row of the i of expression H subgraph, H (i, j+1) the horizontal first order difference value of the capable j+1 row of the i of expression H subgraph, V (i, j) the horizontal first order difference value of the capable j row of the i of expression V subgraph, V (i, j+1) the horizontal first order difference value of the capable j+1 row of the i of expression V subgraph, D (i, j) the horizontal first order difference value of the capable j row of the i of expression D subgraph, D (i, j+1) the horizontal first order difference value of the capable j+1 row of the i of expression D subgraph, Dif_I (i, j) the small echo subgraph first order difference summation of the car plate of the capable j row of expression i;
Step 13. is calculated the global density and the row density of small echo subgraph first order difference summation; Global density is meant the number of nonzero value among the matrix D if_I, its computing method are: counter counter is set, from top to bottom, scan matrix Dif_I from left to right, when running into nonzero value, the value of counter is added 1, and the value among the counter that obtains during been scanned is exactly the value of global density; Row density is meant the non-zero number of each row among the matrix D if_I; The total n row of matrix D if_I, n is the height of image; The computing method of row density are: be provided with n counter counter_1, counter_2 ..., counter_n, counter_1 is used for the nonzero value number of count matrix Dif_I first row, counter_2 is used for the nonzero value number of count matrix Dif_I secondary series, the rest may be inferred, counter_n is used for the nonzero value number of count matrix Dif_I n row, each row among the scan matrix Dif_I from top to bottom, when running into nonzero value, the value of the counter of respective column is added 1, and the value in the individual count device that obtains during been scanned is exactly the row density of each row; At last, the row density matrix pos that 1*n dimension is set preserves the value of row density;
Step 14. is determined the threshold values of row density binaryzation, and binaryzation row density; With 0.2 times global density is that threshold values carries out binaryzation to row density matrix pos, and the method for binaryzation is: the value more than or equal to threshold values among the row density matrix pos is turned to 1 by two-value, is turned to 0 less than the value of threshold values by two-value; Thus, row density matrix pos is the sequence of forming by a series of 0 and 1 by binaryzation; Represent the zone that row density is bigger by the 1 continuous sequence of forming; Represent the zone that row density is less by the 0 continuous sequence of forming; The border in two kinds of zones is represented in 0 to 1 or 1 to 0 saltus step, respectively 0 to 1 and 1 to 0 saltus step location records in storehouse stack3 and stack4;
Step 15. is carried out car plate and is vertically located, and determines the left margin and the right margin of car plate; By the binaryzation of step 14, the continuous pairing position of 1 sequence is license plate area column position in entire image among the row density matrix pos, and the value among the stack3 is the left margin left of license plate area, and the value among the stack4 is the right margin right of license plate area;
Step 16. is determined real license plate area according to the characteristic of car plate; Concrete method is: reject the car plate width less than 30 and the car plate width less than the pseudo-license plate area of car plate height, thereby obtain real license plate area;
Step 17. is utilized four boundary values of the license plate area that step 9 and step 15 obtain, and extracts license plate area; Concrete car plate extraction step is as follows: at first according to the value of car plate coboundary top that obtains in the step 9 and lower boundary bottom, calculate the width plate_width=bottom-top of car plate; Car plate left margin left that obtains according to step 15 and the value of right margin right are again calculated the length plate_length=right-left of car plate; Then, size of definition matrix car_plate that is plate_width*plate_length is used for storing and cuts apart later car plate; At last, give car_plate the capable numerical value assignment capable to bottom, that 1eft is listed as the right row of top in the original image, at this moment, matrix car_plate is automotive license plate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100215409A CN100385450C (en) | 2005-08-26 | 2005-08-26 | Vehicle license plate extraction method based on high-hat switch and wavelet switch |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2005100215409A CN100385450C (en) | 2005-08-26 | 2005-08-26 | Vehicle license plate extraction method based on high-hat switch and wavelet switch |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1920849A CN1920849A (en) | 2007-02-28 |
| CN100385450C true CN100385450C (en) | 2008-04-30 |
Family
ID=37778569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2005100215409A Expired - Fee Related CN100385450C (en) | 2005-08-26 | 2005-08-26 | Vehicle license plate extraction method based on high-hat switch and wavelet switch |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100385450C (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101183425B (en) * | 2007-12-20 | 2010-06-09 | 四川川大智胜软件股份有限公司 | Guangdong and Hong Kong license plate locating method |
| CN101763505B (en) * | 2009-12-29 | 2011-12-07 | 重庆大学 | Vehicle license character feature extracting and classifying method based on projection symmetry |
| CN106326898B (en) * | 2016-08-04 | 2019-09-20 | 安凯 | A kind of license plate image extracting method |
| CN106886777B (en) * | 2017-04-11 | 2020-06-09 | 深圳怡化电脑股份有限公司 | Character boundary determining method and device |
| CN108805127A (en) * | 2017-05-03 | 2018-11-13 | 迅驰（北京）视讯科技有限公司 | Licence plate recognition method, system and car license recognition equipment |
| CN108830274A (en) * | 2018-05-03 | 2018-11-16 | 南京邮电大学 | Characters on license plate detects positioning and recognition methods automatically under a kind of complex background environment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6553131B1 (en) * | 1999-09-15 | 2003-04-22 | Siemens Corporate Research, Inc. | License plate recognition with an intelligent camera |
| CN1529276A (en) * | 2003-10-17 | 2004-09-15 | 沈阳工业学院 | Device and method for automatically identifying registration number for motor vehicle |
-
2005
- 2005-08-26 CN CNB2005100215409A patent/CN100385450C/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6553131B1 (en) * | 1999-09-15 | 2003-04-22 | Siemens Corporate Research, Inc. | License plate recognition with an intelligent camera |
| CN1529276A (en) * | 2003-10-17 | 2004-09-15 | 沈阳工业学院 | Device and method for automatically identifying registration number for motor vehicle |
Non-Patent Citations (4)
| Title |
|---|
| Robust License-Plate Recognition Method for PassingVehicles Under Outside Environment. Takashi Naito, Toshihiko Tsukada, Keiichi Yamada, KazuhiroKozuka, and Shin Yamamoto.IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY,Vol.49 No.6. 2000 |
| Robust License-Plate Recognition Method for PassingVehicles Under Outside Environment. Takashi Naito, Toshihiko Tsukada, Keiichi Yamada, KazuhiroKozuka, and Shin Yamamoto.IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY,Vol.49 No.6. 2000 * |
| 车辆牌照定位的算法研究. 袁志伟，潘晓露，陈艾，李一民.昆明理工大学学报，第26卷第2期. 2001 |
| 车辆牌照定位的算法研究. 袁志伟，潘晓露，陈艾，李一民.昆明理工大学学报，第26卷第2期. 2001 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1920849A (en) | 2007-02-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100414561C (en) | Vehicle plate extracting method based on skiagraphy and mathematical morphology | |
| CN100545858C (en) | Based on the vehicle license plate extraction method in the complex background of wavelet transformation | |
| CN101408942B (en) | Method for locating license plate under a complicated background | |
| Yu et al. | A novel license plate location method based on wavelet transform and EMD analysis | |
| CN102043945B (en) | License plate character recognition method based on real-time vehicle tracking and binary index classification | |
| CN102254152B (en) | License plate location method based on color change points and color density | |
| CN102375982B (en) | Multi-character characteristic fused license plate positioning method | |
| CN100414560C (en) | Vehicle plate extracting method based on small wave conversion and Redon transform | |
| Asif et al. | Multinational vehicle license plate detection in complex backgrounds | |
| CN103198315B (en) | Based on the Character Segmentation of License Plate of character outline and template matches | |
| CN100385450C (en) | Vehicle license plate extraction method based on high-hat switch and wavelet switch | |
| CN105373794A (en) | Vehicle license plate recognition method | |
| CN103136528B (en) | A kind of licence plate recognition method based on dual edge detection | |
| CN103514448A (en) | Method and system for navicular identification | |
| CN103116751A (en) | Automatic license plate character recognition method | |
| CN109145900A (en) | A kind of licence plate recognition method based on deep learning | |
| CN105354568A (en) | Convolutional neural network based vehicle logo identification method | |
| CN104050450A (en) | Vehicle license plate recognition method based on video | |
| CN103605977A (en) | Extracting method of lane line and device thereof | |
| CN103559507A (en) | Method for detecting traffic signs based on combination of color feature and shape feature | |
| CN101154271A (en) | License plate character segmentation method based on fast area labeling algorithm and license plate large-spacing locating method | |
| CN103530608A (en) | Vehicle type judgment method and vehicle type judgment device | |
| CN107704833A (en) | A kind of front vehicles detection and tracking based on machine learning | |
| CN103279755A (en) | Vehicle bottom shadow characteristics-based rapid license plate positioning method | |
| CN104766344A (en) | Vehicle detecting method based on moving edge extractor |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20080430 Termination date: 20100826 |