CN105005757B - A kind of license plate character recognition method popular based on Grassmann - Google Patents
A kind of license plate character recognition method popular based on Grassmann Download PDFInfo
- Publication number
- CN105005757B CN105005757B CN201510108781.0A CN201510108781A CN105005757B CN 105005757 B CN105005757 B CN 105005757B CN 201510108781 A CN201510108781 A CN 201510108781A CN 105005757 B CN105005757 B CN 105005757B
- Authority
- CN
- China
- Prior art keywords
- license plate
- characters
- character
- matrix
- image
- 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.)
- Active
Links
Abstract
The invention discloses a kind of license plate character recognition method popular based on Grassmann.Belong to field technical field of image processing.Grassmann manifolds are applied to during Recognition of License Plate Characters by the present invention, several similar character pictures are handled simultaneously, and similar all character pictures are formed into a matrix, make full use of the contact between all information of characters on license plate image and similar characters on license plate, obtained matrix is subjected to SVD decomposition, obtain opening into subspace for each characters on license plate, these correspond to the point in Grassmann manifolds into subspace, and Recognition of License Plate Characters problem is converted into the distance problem asked in Grassmann manifolds between points.The present invention is used for Vehicle License Plate Recognition System, the correlation between acquired characters on license plate information and similar character is make use of to greatest extent, with very high discrimination, require lower for the image quality of characters on license plate, applied to having good robustness and accuracy in complex environment.
Description
Technical field
The invention belongs to technical field of image processing, more particularly to Recognition of License Plate Characters.
Background technology
With the development of intelligent transportation system, Vehicle License Plate Recognition System is widely used in every field.It is with digitized map
Based on the technologies such as processing, pattern-recognition, computer vision, the vehicle image or video sequence of shot by camera are entered
Row analysis, obtains the number-plate number of each car, so as to complete identification process.Pass through some subsequent treatments, it is possible to achieve parking
Field toll administration, toll station automatic fee management, the measurement of magnitude of traffic flow Con trolling index, high way super speed automation prison
The functions such as pipe, vehicle location, automobile burglar.
Generally, Vehicle License Plate Recognition System can be divided into three parts：License Plate (obtaining single license plate image), characters on license plate point
Cut, Recognition of License Plate Characters.Whole system it is preceding it is two-part on the basis of how to carry out accurate character recognition, just turn into final shadow
The important problem of acoustic system discrimination.
Currently, there is following problem in Recognition of License Plate Characters：
1st, small size character set.Specific to the automotive number plate standard (GA36-2007, GA804) of China's current, Chinese car plate
Contain Chinese character, English alphabet and Arabic numerals.Chinese character is complicated, and the actual character that obtains has adhesion, dirty
Situations such as damage, English alphabet " J ", " L " easily obscure with " 1 "；" D ", " 0 " and " Q " easily obscure etc..
2nd, it is larger to obtain picture quality otherness.Front-end collection equipment obtains vehicle image, it is understood that there may be interference and geometry
Deformation；Test environment is complicated and changeable, may when natural environment harsh conditions such as Vehicle License Plate Recognition System application sleet sky, mists
Cause car plate defaced, while influence to obtain license plate image quality.
Current Recognition of License Plate Characters algorithm mainly has following several method：
(1) Recognition Algorithm of License Plate based on template matches.Utilize the spies such as the profile of characters on license plate, backbone or peak valley projection
Sign, key point extraction first is carried out to character to be identified, i.e., topological analysis is carried out to obtain the key point of character edge to character, then
The classification extraction characters on license plate feature of character is determined, is matched with standard characters on license plate.But due in actual test environment
Obtain the geometry deformation that license plate image has interference and license plate image, characters on license plate correct recognition rata is low and robustness compared with
Difference.
(2) Recognition Algorithm of License Plate based on SVM.According to the feature of characters on license plate, grader is established, grader is established each
The Sample Storehouse of character, train to obtain the discriminant function of each character by SVM methods.Then according to character relevant position, normalization
Corresponding grader group is sent to, classification results are obtained by discriminant function.But due to being difficult to extract that character can be represented very well
The complexity of correlated characteristic and site environment, characters on license plate also are difficult to accurately split, and character otherness is big.Finally result in correct knowledge
Rate is not low and robustness is poor.
The content of the invention
The present invention is directed to above-mentioned technical problem, discloses a kind of image quality to characters on license plate image and requires low, multiple
There is the license plate character recognition method of good robustness and accuracy in miscellaneous environment.
The license plate character recognition method popular based on Grassmann of the present invention, comprises the following steps：
Step A：Structure opening into subspace per class characters on license plate image pattern：
A-1：N characters on license plate image pattern is taken to every class characters on license plate, each characters on license plate image is converted into gray-scale map
Picture, and gray level image is normalized to identical image size, wherein n is more than or equal to 3；
A-2：Per the image array of class n sample of characterWherein i is character types mark
Know symbol,Represent the characters on license plate image of j-th of sample of the i-th class character row wise or column wise
Deploy the m dimensional vectors formed, wherein m represents the pixel sum of the characters on license plate image after normalization, j=1,2 ..., n；
It is describedRepresent the pixel value of p-th of pixel of the characters on license plate image of j-th of sample of the i-th class character, p=1,2 ...,
m；
A-2：Per the image array of class n sample of characterWherein i is character types mark
Know symbol,Represent the characters on license plate image of j-th of sample of the i-th class character row wise or column wise
Deploy the m dimensional vectors formed, j=1,2 ..., n；WhereinRepresent the characters on license plate image of j-th of sample of the i-th class character
P-th of pixel pixel value, p=1,2 ..., m
A-3：To image array MiSingular value SVD decomposition is carried out, is obtainedWherein matrix UiFor m*m ranks
Unitary matrice, matrix AiFor n*n rank diagonal matrix, matrixFor n*n rank unitary matrice ViAssociate matrix；Take diagonal matrix Ai
In preceding k maximum correspondence position matrix ViIn column vector form matrix Yi, with span (Yi) representing matrix YiZhang Chengzi
Space, described it will open into subspace span (Yi) be mapped in Grassmann manifolds, wherein subscript i is character types identifier,
Wherein 1≤k ＜ n；
Step B：Character recognition is carried out to characters on license plate image to be identified
B-1：N similar characters on license plate images to be identified are inputted, character types x to be identified is obtained based on step A-1, A-2
Image arrayBased on step A-3 by image array MxObtain matrix Yx, with span (Yx) table
Show matrix YxInto subspace, will described into subspace span (Yx) be mapped in Grassmann manifolds；
B-2：Based on the distance between two points that Grassmann is popular, calculate open into subspace span (Y respectivelyx) and each class
Into subspace span (Yi) between point distance, it apart from corresponding type identifier i is character to be identified to take smallest point
Type x character species.
Beneficial effect：Traditional Recognition of License Plate Characters mode is that the character of each segmentation is handled, and forms feature
The problem of set, model is established with characteristic set, and the method is present：For same class characters on license plate, each car plate word has been isolated
Correlation between symbol, no matter using which kind of feature extraction mode, it can all cause the loss of primitive character or increase useless
Feature.And the present invention uses the license plate character recognition method based on Grassmann manifolds, by multiple character set of same character types
It is fated into a matrix, each column vector of matrix is a secondary characters on license plate image, is the advantages of the processing mode, it is contemplated that
With the correlation between the characters on license plate image of character types, every width characters on license plate image direct vectorization is become into vector, kept away
The loss of any characteristic information is exempted from, make use of existing information to carry out Recognition of License Plate Characters to the full extent, greatly improve
The accuracy and robustness of identification.
Brief description of the drawings
The present invention will illustrate by embodiment and with reference to the appended drawing, wherein：
Fig. 1 is in embodiment, to the identification process figure of characters on license plate image to be identified.
Embodiment
To make the object, technical solutions and advantages of the present invention clearer, with reference to embodiment and accompanying drawing, to this hair
It is bright to be described in further detail.
To make the object, technical solutions and advantages of the present invention clearer, with reference to embodiment and accompanying drawing, to this hair
It is bright to be described in further detail.
The present invention is used for the License Plate Character Segmentation of Vehicle License Plate Recognition System.Vehicle License Plate Recognition System receives what monitoring device collected
Image data stream carries out License Plate, to obtain single license plate image.Currently, obtaining the usual processing mode of license plate image is：
Method based on mixed Gaussian background modeling, the foreground and background in moving scene is obtained using weights and variance, will currently be obtained
A two field picture and background image subtraction can be to obtain each moving vehicle of the motion target area i.e. in scene.Then root
According to each vehicle condition tracked in scene, first obtained single-frame images is carried out being converted into gray level image, and to gray-scale map
As carrying out rim detection, binaryzation is carried out again to the edge-detected image of acquisition, can so remove obvious interference and some
The influence of noise spot caused by noise, especially night car light, based on pre-structured matrix, (row and column of matrix can basis afterwards
The length-width ratio of car plate is set, and the element in matrix is initialized as 1) traveling through whole target area, then extracts connected domain, then
Morphology closed operation is carried out to obtained target area, each connected domain is demarcated afterwards and seeks its minimum external square
Shape, then, obtain boundary rectangle corresponding image in artwork.Finally, obtained image is classified and obtains car plate
Positive negative sample, some features that car plate has are chosen, remove pseudo- car plate using the grader of positive and negative sample training two of car plate, so as to
Obtain the license plate image of coarse positioning.Car plate is further accurately positioned, obtains the license plate image of fine positioning：First, to coarse positioning
License plate image projected in the horizontal direction, the up-and-down boundary of car plate is accurately determined, then to car plate in the vertical direction
Projection, threshold values is selected, accurate location of the candidate license plate region as car plate left and right edges is judged according to threshold values, it is fixed so as to obtain essence
The license plate image of position.
Output obtain each license plate image Character segmentation position when, can be existing either method or
The present inventor is in application title《A kind of registration number character dividing method of dynamic template combination pixel》Described in
Dividing method, i.e., binary conversion treatment is carried out to the license plate image of input first, and calculate the license plate image after binary conversion treatment
Width w and height h；Then plate template is set：Height and license plate image (license plate image after binary conversion treatment) to be split
Height it is identical, be arranged to h, width is arranged to w'(initial values and is arranged to w/3~2w/3), plate template set 7 characters, respectively
The ratio of width to height of character is arranged to r1, the ratio of width to height at the 2nd interval between the 3rd character is r2, the interval of other intercharacters
The ratio of width to height be r3, wherein the ratio of width to height r1, r2, r3 value based on automotive number plate standard (such as《The People's Republic of China (PRC)
Industry standards of public safety-People's Republic of China's automotive number plate》(GA36-2007)) in defined correspondingly-sized carry out
Set；Then template slip processing is performed：Using the left end of the license plate image after binary conversion treatment as original position, based on default
Plate template carries out template slip, slides a pixel every time, often slides once, then records license plate image in current car plate mould
Split position corresponding to the number of non-zero pixels point at each character position of plate and current plate template, when plate template is right
When end reaches the right-hand member of the license plate image or the original position of plate template slides into the T of license plate image (T value is w/
4~w/2) place when terminate to slide；Gradually the width w'(of increase plate template highly keeps constant), keep each character of plate template
And the ratio of width to height (r1, r2, r3) of intercharacter space, template slip processing is repeated, until the width w' increases of plate template
To the width w of license plate image, width w' increase step-length is arranged to 1~3 pixel；Processing when institute is slided based on each template
The number of the non-zero pixels point of record, search comprising the characters on license plate the largest number of split positions of pixel as license plate image
Split position.
Character recognition is carried out to the characters on license plate image to be identified of acquisition：
(1) opening into subspace per class characters on license plate image (known characters on license plate) sample is built：
Step (1)：The individual characters on license plate image patterns of n (n >=3) are taken to every class characters on license plate, each characters on license plate image is turned
Gray level image is turned to, and is m=w*h by gray level image normalization size, w represents the width of characters on license plate image, and h represents car plate
The height of character picture, w, h disposal ability of the value based on system are set, and present embodiment is based on standard size
Set, i.e. w=45, h=90；
Step (2)：Image array will be expressed as per a kind of characters on license plate image normalization：(wherein subscript i is used to identify different character types marks in the present invention, such as different
Numeral, different English alphabets, different men corresponds to a kind of character types respectively), wherein(j=1,2 ..., n) table
Show that the characters on license plate image of j-th of sample of the i-th class character deploys the m dimensional vectors to be formed row wise or column wise, i.e.,Wherein(p=1,2 ..., m) represents the car plate of j-th of sample of the i-th class character
The pixel value of p-th of pixel of character picture, i.e. the image array M in m*n ranksiIn, each column vector corresponds to a width respectively
Characters on license plate image, every width characters on license plate image direct vectorization is become into vectorExtract and realize with existing feature based
The processing mode of character recognition is compared, and can effectively avoid the loss of any characteristic information.
Step (3)：To image array MiSVD decomposition is carried out, is obtainedWherein matrix UiFor m*m ranks
Unitary matrice, matrix AiFor n*n rank diagonal matrix, matrixFor n*n rank unitary matrice ViAssociate matrix.Because m is much big
In n, to reduce amount of calculation, diagonal matrix A is taken hereiniIn the individual maximum correspondence position of preceding k (1≤k ＜ n) matrix ViIn row to
Amount forms matrix Yi.Such as diagonal matrixCurrent setting k value is 2, then AiIn preceding k maximum
Correspondence position is that the 1st row 1 arranges, the 4th row the 4th row, then by matrix ViIn the 1st row, the column vector of the 4th row form matrix Yi。
Then, with span (Yi) representing matrix YiInto subspace (also referred to as generated subspace), by this into subspace
span(Yi) be mapped in Grassmann manifolds (and Grassmann (Jim Glassman) manifold be on set of real numbers m dimension linear son
Space).Because corresponding one of each point in Grassmann manifolds is opened into subspace, then the identification to characters on license plate is just
It is converted to and calculates the distance between two points different in Grassmann manifolds problem.
(2) opening into subspace based on every class characters on license plate image pattern enters line character to characters on license plate image to be identified
Identification, it is as follows referring to Fig. 1, concrete processing procedure：
N similar characters on license plate images to be identified are inputted, video can be collected based on the monitoring device of Vehicle License Plate Recognition System
In n two field pictures in stream, n license plate image of same car plate is got, then based on existing any license plate image dividing method institute
The Character segmentation position of determination, obtain the same character types of n (in n width license plate images, the same character position of same car plate
Character) characters on license plate image.
Current character types to be identified are represented with x, based on structure opening into subspace per class characters on license plate image pattern
Step (1), (2) identical building mode, build the image array on n width characters on license plate imagesI.e. after gray proces and size (m=w*h) normalization, by n width characters on license plate image by row
Or the m dimensional vectors for deploying to be formed by row(j=1,2 ..., n) form image array
Mx。
Then, the opening into (3) identical building mode the step of subspace per class characters on license plate image pattern based on structure,
Build correspondence image matrix MxInto subspace span (Yx), i.e., to image array MxSVD decomposition is carried out, is obtainedMatrix UxFor the unitary matrice of m*m ranks, matrix AxFor n*n rank diagonal matrix, matrixFor n*n rank tenth of the twelve Earthly Branches squares
Battle array ViAssociate matrix.Take diagonal matrix AxIn preceding k maximum correspondence position matrix VxIn column vector form matrix
Yx.With span (Yx) representing matrix YxInto subspace, and by span (Yx) be mapped in Grassmann manifolds.
Then, with known characters on license plate image into subspace span (Yi) ask point in Grassmann manifolds away from
From characters on license plate image to be identified is identified.Concrete processing procedure is：
Two are calculated into subspace (span (Yx) and span (Yi)) diagonal matrix cos θ, by (Yi)H(Yx) carry out SVD
Decompose, i.e. (Yi)H(Yx)=Ucos θ VH, obtain diagonal matrix cos θ=diag (cos θ1,cosθ2,…,cosθn), according to
Point span (Y in Grassmann manifoldsi) and point span (Yx) point distance d (Yi,Yx) two can be calculated into subspace
(span(Yx) and span (Yi)) based on the point distance in Grassmann manifolds：
Finally, according to Nearest neighbor rule, minimum range (min (d (Y are takeni,Yx))) corresponding to known characters on license plate classification i make
For the result of characters on license plate to be identified.
Claims (1)
1. based on the license plate character recognition method that Grassmann is popular, it is characterised in that comprise the following steps：
Step A：Structure opening into subspace per class characters on license plate image pattern：
A-1：N characters on license plate image pattern is taken to every class characters on license plate, each characters on license plate image is converted into gray level image, and
Gray level image is normalized to identical image size, wherein n is more than or equal to 3；
A-2：Per the image array of class n sample of characterWherein i is character types identifier,Represent that the characters on license plate image of j-th of sample of the i-th class character deploys row wise or column wise
The m dimensional vectors of formation, wherein m represent the pixel sum of the characters on license plate image after normalization, j=1,2 ..., n；Table
Show the pixel value of p-th of pixel of the characters on license plate image of j-th of sample of the i-th class character, p=1,2 ..., m；
A-3：To image array MiSingular value SVD decomposition is carried out, is obtainedWherein matrix UiFor m*m rank tenth of the twelve Earthly Branches squares
Battle array, matrix AiFor n*n rank diagonal matrix, matrix Vi HFor n*n rank unitary matrice ViAssociate matrix；Take diagonal matrix AiIn preceding k
The matrix V of individual maximum correspondence positioniIn column vector form matrix Yi, with span (Yi) representing matrix YiInto subspace,
Described it will open into subspace span (Yi) be mapped in Grassmann manifolds, wherein subscript i is character types identifier, wherein 1
≤ k ＜ n；
Step B：Character recognition is carried out to characters on license plate image to be identified：
B-1：N similar characters on license plate images to be identified are inputted, character types x to be identified figure is obtained based on step A-1, A-2
As matrixBased on step A-3 by image array MxObtain matrix Yx, with span (Yx) represent square
Battle array YxInto subspace, will described into subspace span (Yx) be mapped in Grassmann manifolds；
B-2：Based on the distance between two points that Grassmann is popular, calculate open into subspace span (Y respectivelyx) with each class into
Subspace span (Yi) between point distance, it apart from corresponding type identifier i is character types x to be identified to take smallest point
Character species.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510108781.0A CN105005757B (en) | 2015-03-12 | 2015-03-12 | A kind of license plate character recognition method popular based on Grassmann |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510108781.0A CN105005757B (en) | 2015-03-12 | 2015-03-12 | A kind of license plate character recognition method popular based on Grassmann |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105005757A CN105005757A (en) | 2015-10-28 |
| CN105005757B true CN105005757B (en) | 2018-04-06 |
Family
ID=54378423
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510108781.0A Active CN105005757B (en) | 2015-03-12 | 2015-03-12 | A kind of license plate character recognition method popular based on Grassmann |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105005757B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550645B (en) * | 2015-12-08 | 2019-02-12 | 北京工业大学 | A kind of least squared classified method in product Grassmann manifold |
| CN109034173A (en) * | 2017-06-08 | 2018-12-18 | 北京君正集成电路股份有限公司 | Target object choosing method and device |
| CN111507243B (en) * | 2020-04-15 | 2022-08-19 | 桂林电子科技大学 | Human behavior recognition method based on Grassmann manifold analysis |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101872416B (en) * | 2010-05-06 | 2013-05-01 | 复旦大学 | Vehicle license plate recognition method and system of road image |
| CN103226696A (en) * | 2013-04-07 | 2013-07-31 | 布法罗机器人科技（苏州）有限公司 | License plate recognition system and license plate recognition method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9092979B2 (en) * | 2010-12-14 | 2015-07-28 | Xerox Corporation | Automated license plate recognition system and method using human-in-the-loop based adaptive learning |
-
2015
- 2015-03-12 CN CN201510108781.0A patent/CN105005757B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101872416B (en) * | 2010-05-06 | 2013-05-01 | 复旦大学 | Vehicle license plate recognition method and system of road image |
| CN103226696A (en) * | 2013-04-07 | 2013-07-31 | 布法罗机器人科技（苏州）有限公司 | License plate recognition system and license plate recognition method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105005757A (en) | 2015-10-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105046196B (en) | Front truck information of vehicles structuring output method based on concatenated convolutional neutral net | |
| CN101398894B (en) | Automobile license plate automatic recognition method and implementing device thereof | |
| CN104715252B (en) | A kind of registration number character dividing method of dynamic template combination pixel | |
| CN104134079B (en) | A kind of licence plate recognition method based on extremal region and extreme learning machine | |
| CN103810505B (en) | Vehicles identifications method and system based on multiple layer description | |
| CN103268489B (en) | Automotive number plate recognition methods based on sliding window search | |
| Dehghan et al. | View independent vehicle make, model and color recognition using convolutional neural network | |
| CN106683119B (en) | Moving vehicle detection method based on aerial video image | |
| Gomez et al. | Traffic lights detection and state estimation using hidden markov models | |
| CN106529532A (en) | License plate identification system based on integral feature channels and gray projection | |
| CN105718912B (en) | A kind of vehicle characteristics object detecting method based on deep learning | |
| CN103310194A (en) | Method for detecting head and shoulders of pedestrian in video based on overhead pixel gradient direction | |
| Soomro et al. | Vehicle number recognition system for automatic toll tax collection | |
| CN103279738B (en) | Automatic identification method and system for vehicle logo | |
| CN104200207A (en) | License plate recognition method based on Hidden Markov models | |
| CN108509950B (en) | Railway contact net support number plate detection and identification method based on probability feature weighted fusion | |
| CN105718866A (en) | Visual target detection and identification method | |
| Daramola et al. | Automatic vehicle identification system using license plate | |
| CN102147867A (en) | Method for identifying traditional Chinese painting images and calligraphy images based on subject | |
| CN105005757B (en) | A kind of license plate character recognition method popular based on Grassmann | |
| CN110516527B (en) | Visual SLAM loop detection improvement method based on instance segmentation | |
| CN104268509B (en) | The method and system of dump truck car plate detection | |
| Lee et al. | Real-time automatic vehicle management system using vehicle tracking and car plate number identification | |
| Adytia et al. | Indonesian license plate detection and identification using deep learning | |
| CN111079543A (en) | Efficient vehicle color identification method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210517 Address after: No.3, 11th floor, building 6, no.599, shijicheng South Road, Chengdu hi tech Zone, China (Sichuan) pilot Free Trade Zone, Chengdu, Sichuan 610041 Patentee after: Houpu clean energy Co.,Ltd. Address before: 611731, No. 2006, West Avenue, Chengdu hi tech Zone (West District, Sichuan) Patentee before: University of Electronic Science and Technology of China |
|
| TR01 | Transfer of patent right | ||
| CP01 | Change in the name or title of a patent holder |
Address after: No.3, 11th floor, building 6, no.599, shijicheng South Road, Chengdu hi tech Zone, China (Sichuan) pilot Free Trade Zone, Chengdu, Sichuan 610041 Patentee after: Houpu clean energy (Group) Co.,Ltd. Address before: No.3, 11th floor, building 6, no.599, shijicheng South Road, Chengdu hi tech Zone, China (Sichuan) pilot Free Trade Zone, Chengdu, Sichuan 610041 Patentee before: Houpu clean energy Co.,Ltd. |
|
| CP01 | Change in the name or title of a patent holder |