CN103065137A - License plate character recognition method - Google Patents

Abstract

The invention provides a novel license plate character recognition method. The method of combining the canny algorithm and binary images is adopted for recognizing characters. The edges and hop information of the characters are extracted in the process of recognition, hop of edge pixels of the characters are matched with hop of a template character set, a template character with the highest matched-degree is discovered, and then recognition results of the characters are further obtained. The character recognition method based on the hop solves character recognition problems under various kinds of interference well and overcomes various interference factors effectively, and a stable high-recognition rate can be maintained. In addition, the standard canny algorithm is simplified, recognition efficiency is improved greatly, occupied system resources are reduced, and recognition speed is accelerated.

Description

A kind of license plate character recognition method

Affiliated technical field:

Patent of the present invention relates to a kind of character identifying method in the Vehicle License Plate Recognition System, belongs to technical field of image processing.

Background technology:

Along with the fast development of Chinese national economy, the requirement of control of traffic and road is progressively improved, intelligent transportation system is arisen at the historic moment.In intelligent transportation system, after accurate positioning licence plate and Character segmentation, carry out character recognition by Vehicle License Plate Recognition System and just finally finish complete car plate identification, so the quality of character recognition plays vital effect to character identification rate.Therefore, as the important component part of intelligent transportation system, Vehicle License Plate Recognition System has obtained fast development, by the Chinese scholars broad research.Vehicle License Plate Recognition System is divided into Image Acquisition, car plate location, Character segmentation, character recognition four parts, and wherein character recognition is the Research Emphasis of each enterprise of the industry.Aspect the character feature extraction, need to suppress various interference, rotation and the deformation of, character even such as uneven illumination, stroke weight is inconsistent and the situations such as fracture or adhesion, just can extract stable character feature, finally finishes character recognition.

Common threshold obtains bianry image in the prior art, and discrimination is not high when light is inhomogeneous, and original algorithm steps is more, and the system resource that takies when identifying is too high, and recognition speed is not high; Simultaneously, for license plate sloped, characters on license plate stroke fracture, character are stained because turn inside diameter, shooting angle etc. cause, low light according to the time noise etc., all can cause discrimination to descend.

Summary of the invention:

In order to improve the discrimination of character, the present invention proposes a kind of method of new Recognition of License Plate Characters, adopt the method for canny algorithm and bianry image combination to come identification character, and the canny algorithm of standard is simplified, specific as follows:

At first license plate grey level image is divided into some character zones, to the gray level image I of each character zone _CharOrder is processed in accordance with the following steps:

A, to I _CharAsk adaptive threshold;

B, to I _CharAdopt the Canny algorithm to determine the character edge image I _Edge

C, to above-mentioned character edge image I _EdgeCarry out filling cavity;

D, again determine above-mentioned character edge image I _EdgeThe border;

E, to above-mentioned I _EdgeCarry out normalized;

The horizontal hopping sequences S at F, calculating character edge _HoriWith vertical transition sequence S _Vert, and according to above-mentioned horizontal hopping sequences S _HoriWith vertical transition sequence S _VertDetermine upper number of transitions density Hop_Density _Top, lower number of transitions density Hop_Density _Bottom, left number of transitions density Hop_Density _Left, right number of transitions density Hop_Density _Right, and further determine up and down number of transitions density ratio Hop_Density _Top/BottomWith left and right sides number of transitions density ratio Hop_Density _Left/Right

G, with the above-mentioned horizontal hopping sequences S that determines in the above-mentioned F step _Hori, vertical transition sequence S _Vert, number of transitions density ratio Hop_Density up and down _Top/BottomWith left and right sides number of transitions density ratio Hop_Density _Left/RightWith the horizontal hopping sequences SM that concentrates the current character template that reads from Character mother plate _Hori, vertical transition sequence SM _Vert, number of transitions density ratio Hop_Density_M up and down _Top/Bottom, left and right sides number of transitions density ratio Hop_Density_M _Left/RightMate one by one, find the highest character index value of matching degree, finish character recognition.

Adopt method of the present invention that the characters on license plate gray level image is processed, owing to extracted edge and the saltus step information of character, again coupling is done in the saltus step of character edge pixel and the saltus step of template character set, found the highest template character of matching degree, and then obtain the recognition result of character.This character recognition based on saltus step has solved character identification problem under the various interference preferably, can effectively overcome various disturbing factors, keeps comparatively stable high discrimination.And, because the canny algorithm of standard is simplified, greatly improved the efficient of identification, reduced the system resource that takies, improved recognition speed.

Embodiment:

Method of the present invention is at first license plate grey level image to be carried out Character segmentation, and license plate image is divided into several character zones, to the gray level image I of each character zone _CharRepeat following operation, finally obtain the index value Index that each character is concentrated at predetermined template character _m, finish identification to all characters with this.Wherein, to be that the situations such as numeral, letter, Chinese character, font are self-defined (for example, can comprise each province's abbreviation, 26 English alphabet capital and small letters to the characters on license plate that predetermined masterplate character set can be identified as required, numeral can also be set other literal or other language as required).Specifically to gray level image I _CharTreatment step as follows:

1. to I _CharAsk adaptive threshold T

2. to I _CharAdopt the Canny algorithm to obtain the character edge image I _Edge

1) adopt the template of 3*3, ask about the current pixel point P, about, the gradient G of two clinodiagonals ₁, G ₂, G ₃, G ₄, the corresponding operator of each gradient direction is respectively H ₁, H ₂, H ₃, H ₄:

$H_1=\left[\begin{array}{ccc}0& -1& 0\\ 0& 2& 0\\ 0& -1& 0\end{array}\right]$ $H_2=\left[\begin{array}{ccc}0& 0& 0\\ -1& 2& -1\\ 0& 0& 0\end{array}\right]$ $H_3=\left[\begin{array}{ccc}-1& 0& 0\\ 0& 2& 0\\ 0& 0& -1\end{array}\right]$ $H_4=\left[\begin{array}{ccc}0& 0& -1\\ 0& 2& 0\\ -1& 0& 0\end{array}\right]$

2) find maximum gradient G _Max=MAX4 (G ₁, G ₂, G ₃, G ₄), and with following mode mark P point:

Namely according to G _Max=MAX4 (G ₁, G ₂, G ₃, G ₄) find the greatest gradient of current pixel point P, and judge current pixel point P whether satisfy simultaneously greatest gradient greater than the gray-scale value of predetermined minimum rim value and current pixel point whether greater than above-mentioned adaptive threshold, if it is mark is judged to be marginal point, I _EdgeBe labeled as 1; Not marginal point, I if not then being judged to be _EdgeBe labeled as 0.

Gray (P) gray-scale value of ordering for P wherein, MIN_EDGE_GRADIENT is defaulted as 15, can get different values according to the difference of captured image type.

3) get back to step 1), repeat this process, until travel through completely, finally obtain edge image I _Edge

3. to the character edge image I _EdgeFilling cavity is with the impact of cancellation on character saltus step calculating.Be specially

1) at I _EdgeIn, obtain current pixel point P _Ij

2) if Gray is (P _Ij)==0 is then with P _IjCentered by get one 3 * 3 window, pixel value consists of matrix W in the window _{3 * 3}, the definition operator $H=\left[\begin{array}{ccc}1& 1& 1\\ 1& 0& 1\\ 1& 1& 1\end{array}\right],$ Ask the scalar product C of the two _Ij=W _{3 * 3}H works as C _Ij〉=T _c(T _c=6), just at I _EdgeMiddle with P _IjBe set to marginal point.

3) get back to step 1, repeat this process, until travel through complete.

4. recomputate the border of character edge image, the purpose of doing like this is in order to eliminate the impact of noise at the boundary

1) looks for I _EdgeUpper marginal position index Index _Top

Method: row travels through from top to bottom, the Num until count in the row edge _Hori〉=T _Num, T _Num=1;

2) look for I _EdgeLower limb location index Index _Bottom

Method: row travels through from top to bottom, the Num until count in the row edge _Hori〉=T _Num, T _Num=1;

3) look for I _EdgeLeft hand edge location index Index _Left

Method: from left to right row travel through, until column border points N um _Vert〉=T _Num, T _Num=1;

4) look for I _EdgeRight hand edge location index Index _Right

Method: row travel through from right to left, until column border points N um _Vert〉=T _Num, T _Num=1.

5. to I _EdgeDo normalized, to reduce the requirement to the Character segmentation precision

1) Index _Top, Index _Bottom, index _Left, Index _RightFour determined images in border are I _{Divided_edge}, with I _{Divided_edge}Do size normalization, its wide height is equated with the wide height of Character mother plate, obtain I _{Normal_edge}

2) record I _{Normal_edge}Height H eight, width is Width.

3)

6. the horizontal and vertical hopping sequences S at calculating character edge _HoriAnd S _Vert, and obtain up and down number of transitions density Hop_Density _Top, Hop_Density _Bottom, left and right sides number of transitions density Hop_Density _Left, Hop_Density _Right, number of transitions density ratio Hop_Density up and down _Top/BottomWith left and right sides number of transitions density ratio Hop_Density _Left/Right

1) obtains horizontal hopping sequences S _Hori

Method: row travels through I from top to bottom _{Normal_edge}All row are for the capable R of j _jIf there is some P _i(P _I-1=0, P _i=1), R then _jOn number of transitions Hop _jAdd 1(Hop _jInitial value is 0), traveled through so all row (R ₁, R ₂.., R _j..., R _Height), obtain horizontal hopping sequences

S _Hori={Hop ₁,Hop ₂，...,Hop _j,...,Hop _Height}；

2) obtain vertical transition sequence S _Vert

Method: from left to right row travel through I _{Normol_edge}All row are for j row C _jIf there is some P _i(p _I-1=0, P _i=1), C then _jOn number of transitions Hop _jAdd 1(Hop _jInitial value is 0), traveled through so all row (C ₁, C ₂..., C _j..., C _Width), obtain vertical transition sequence S _Vert={ Hop ₁, Hop ₂..., Hop _j..., Hop _Width;

3) obtain up and down number of transitions density Hop_Density _Top, Hop_Density _Bottom

$\mathrm{Hop}\_{\mathrm{Density}}_{\mathrm{Top}}=\frac{1}{k}\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{Hop}}_j$ (the k default value is Height/2, Hop _j∈ S _Hori)

${\mathrm{Hop}\_\mathrm{Density}}_{\mathrm{Bottom}}=\frac{1}{h-k}\underset{j=k+1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{Hop}}_j$

(the k default value is Height/2, h=Height, Hop _j∈ S _Hori)

4) obtain left and right sides number of transitions density Hop_Density _Left, Hop_Density _Right

$\mathrm{Hop}\_{\mathrm{Density}}_{\mathrm{Left}}=\frac{1}{k}\underset{j=1}{\overset{k}{\mathrm{\Σ}}}{\mathrm{Hop}}_j$ (k gets default value Width/2, Hop _j∈ S _Vert)

${\mathrm{Hop}\_\mathrm{Density}}_{\mathrm{Right}}=\frac{1}{h-k}\underset{j=k+1}{\overset{h}{\mathrm{\Σ}}}{\mathrm{Hop}}_j$

(k gets default value Width/2, h=Width, Hop _j∈ S _Vert)

5) number of transitions density ratio up and down:

Hop_Density _Top/Bottom=Hop_Density _Top/Hop_Density _Bottom

Left and right sides number of transitions density ratio:

Hop_Density _Left/Right=Hop_Density _Left/Hop_Density _Rigth

7, coupling done one by one in all characters of Character mother plate collection, finds the highest Character mother plate index value of matching degree

1) at the concentrated current character template Tamplate that obtains of Character mother plate _i(1≤i≤n, n are the template set maximum index value) obtains its horizontal hopping sequences SM _Hori, vertical transition sequence SM _Vert, number of transitions density ratio Hop_Density_M up and down _Top/Bottom, left and right sides number of transitions density ratio Hop_Density_M _Left/Right

2) S _HoriWith SM _HoriDo coupling, obtain matching degree M ₁

3) S _VertWith SM _VertDo coupling, obtain matching degree M ₂

4) Hop_Density _Top/BottomWith Hop_Density_M _Top/BottomDo coupling, obtain mating M ₃

5) Hop_Density _Left/RightWith Hop_Density_M _Left/RightDo coupling, obtain matching degree M ₄

6) calculate total matching degree Mt _i: Mt _i=w ₁M ₁+ w ₂M ₂+ w ₃M ₃+ w ₄M ₄Wherein, w ₁, w ₂, w ₃, w ₄Be respectively M ₁, M ₂, M ₃, M ₄Weight

7) get back to step 1, repeat this process, until travel through complete;

8) get Index _mBe { Mt ₁, Mt ₂..., Mt _i..., Mt _nIn peaked index value, Index _mBe final recognition result.So far, this character recognition is complete.

Claims (5)

1. a license plate character recognition method is divided into some character zones with license plate grey level image, to the gray level image I of each character zone _CharOrder is processed in accordance with the following steps:

A, to I _CharAsk adaptive threshold;

B, to I _CharAdopt the Canny algorithm to determine the character edge image I _Edge

C, to above-mentioned character edge image I _EdgeCarry out filling cavity;

D, again determine above-mentioned character edge image I _EdgeThe border;

E, to above-mentioned I _EdgeCarry out normalized;

The horizontal hopping sequences S at F, calculating character edge _HoriWith vertical transition sequence S _Vert, and according to above-mentioned horizontal hopping sequences S _HoriWith vertical transition sequence S _VertDetermine upper number of transitions density Hop_Density _Top, lower number of transitions density Hop_Density _Bottom, left number of transitions density Hop_Density _Left, right number of transitions density Hop_Density _Right, and further determine up and down number of transitions density ratio Hop_Density _Top/BottomWith left and right sides number of transitions density ratio Hop_Density _Left/Right

G, with the above-mentioned horizontal hopping sequences S that determines in the above-mentioned F step _Hori, vertical transition sequence S _Vert, number of transitions density ratio Hop_Density up and down _Top/BottomWith left and right sides number of transitions density ratio Hop_Density _Left/RightWith the horizontal hopping sequences SM that concentrates the current character template that reads from Character mother plate _Hori, vertical transition sequence SM _Vert, number of transitions density ratio Hop_Density_M up and down _Top/Bottom, left and right sides number of transitions density ratio Hop_Density_M _Left/RightMate one by one, find the highest character index value of matching degree, finish character recognition.

2. license plate character recognition method as claimed in claim 1 further is included in and adopts the canny algorithm to obtain the character edge image among the above-mentioned steps B, is specially:

B1, adopt the template of 3*3, ask about the current pixel point P, about, the gradient G of two clinodiagonals ₁, G ₂, G ₃, G ₄, the corresponding operator of each gradient direction is respectively H ₁, H ₂, H ₃, H ₄:

$H_1=\left[\begin{array}{ccc}0& -1& 0\\ 0& 2& 0\\ 0& -1& 0\end{array}\right]$ $H_2=\left[\begin{array}{ccc}0& 0& 0\\ -1& 2& -1\\ 0& 0& 0\end{array}\right]$ $H_3=\left[\begin{array}{ccc}-1& 0& 0\\ 0& 2& 0\\ 0& 0& -1\end{array}\right]$ $H_4=\left[\begin{array}{ccc}0& 0& -1\\ 0& 2& 0\\ -1& 0& 0\end{array}\right]$

According to G _Max=MAX4 (G ₁, G ₂, G ₃, G ₄) find the greatest gradient of current pixel point P, and judge current pixel point P whether satisfy simultaneously greatest gradient greater than the gray-scale value of predetermined minimum rim value and P whether greater than above-mentioned adaptive threshold, if it is mark is judged to be marginal point, I _EdgeBe labeled as 1; Not marginal point, I if not then being judged to be _EdgeBe labeled as 0;

B2, get back to step B2, repeat this process, until travel through completely, finally obtain edge image I _Edge

3. license plate character recognition method as claimed in claim 2 further comprises among the above-mentioned steps C the character edge image I _EdgeThe method of filling cavity is:

C1, at I _EdgeIn, obtain current pixel point P _Ij

If C2 Gray is (P _Ij)==0 is then with P _IjCentered by get one 3 * 3 window, pixel value consists of matrix W in the window _{3 * 3}, the definition operator $H=\left[\begin{array}{ccc}1& 1& 1\\ 1& 0& 1\\ 1& 1& 1\end{array}\right],$ Ask the scalar product C of the two _Ij=W _{3 * 3}H works as C _Ij〉=T _c(T _c=6), just at I _EdgeMiddle with P _IjBe set to marginal point.

C3, get back to step C1, repeat this process, until travel through complete.

4. characters on license plate as claimed in claim 3 sets method, further is included in again to determine above-mentioned character edge image I among the above-mentioned steps D _EdgeThe method on border be:

D1, row traversal to be to determine I from top to bottom _EdgeUpper marginal position index Index _Top

D2, row traversal to be to determine I from top to bottom _EdgeLower limb location index Index _Bottom

D3, from left to right row traversal to be to determine I _EdgeLeft hand edge location index Index _Left

D4, row traversal to be to determine I from right to left _EdgeRight hand edge location index Index _Right

5. want the described license plate character recognition method of one of 1-4 such as right, comprise that further the method with Character mother plate collection coupling is as follows among the above-mentioned steps G:

1) at the concentrated current character template Tamplate that obtains of Character mother plate _i, obtain its horizontal hopping sequences SM _Hori, vertical transition sequence SM _Vert, number of transitions density ratio Hop_Density_M up and down _Top/Bottom, left and right sides number of transitions density ratio Hop_Density_M _Left/Right

2) S _HoriWith SM _HoriDo coupling, obtain matching degree M ₁

3) S _VertWith SM _VertDo coupling, obtain matching degree M ₂

4) Hop_Density _Top/BottomWith Hop_Density_M _Top/BottomDo coupling, obtain mating M ₃

5) Hop_Density _Left/RightWith Hop_Density_M _Left/RightDo coupling, obtain matching degree M ₄

6) calculate total matching degree Mt _i: Mt _i=w ₁M ₁+ w ₂M ₂+ w ₃M ₃+ w ₄M ₄

Wherein, w ₁, w ₂, w ₃, w ₄Be respectively M ₁, M ₂, M ₃, M ₄Weight, its weight numerical value can be according to concrete image different self-defined.

7) get back to step 1), repeat this process, until travel through complete;

8) get Index _mBe { Mt ₁, Mt ₂..., Mt _i..., Mt _nIn peaked index value, Index _mBe final recognition result.