CN105224962A - A kind of similar vehicle license plate extraction method and device - Google Patents

Abstract

The invention discloses a kind of similar vehicle license plate extraction method and device, appointment car plate is split into character string dimension P by the method _c, obtain character string dimension P _clength N, obtains character string dimension P _cthe first-phase of the similar character that each characters on license plate is corresponding and similar value composition thereof is like character matrix P _ewith the first similarity score matrix E _e; Extract described first similarity score matrix E _ein the maximal value that often arranges, to sort from big to small composition array E by its value _es, according to array E _esarray A is set ₁, A ₂..., A _n; According to described first-phase like character matrix P _ewith the first similarity score matrix E _e, calculate the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified successively, put into array A respectively ₁, A ₂..., A _n, until array A ₁~ A _kmiddle number of elements meets similar car plate specified quantity or calculates end, exports similar car plate.The inventive system comprises car plate analysis module, division module, computing module and output module.Method and apparatus of the present invention calculates similar car plate fast by multiple iteration, and calculated amount is little, has stronger dirigibility and extendability.

Description

A kind of similar vehicle license plate extraction method and device

Technical field

The invention belongs to technical field of data recognition, particularly relate to a kind of similar vehicle license plate extraction method and device.

Background technology

Along with the continuous expansion of construction of high-tech traffic system, government, public security, traffic police and ruck often can utilize license board information to find effective record from crossing car data of magnanimity, realize the quick position of vehicle and search.But license plate recognition technology cannot reach the accuracy of 100% forever on the one hand, the memory of people always also exists the possibility of mistake or omission on the other hand.Therefore many times user often cannot determine that whether the license board information needing to search is correct, by car plate Similarity algorithm, customer analysis can be helped to go out one group of license board information maximum with certain car plate similarity degree, improve the search efficiency of user.

Similarity of character string algorithm (LevenshteinDistance) is often utilized to calculate smallest edit distance between two car plates in prior art, namely calculating and convert minimum editing operation number of times needed for another character string to by a character string, finding out one group of the most similar license board information by traveling through all car plate data.LevenshteinDistance algorithm calculates smallest edit distance to known two car plates, for the one group of license board information finding the specific car plate similarity degree with certain maximum, need could be realized by the mode traveling through all car plates, China's car plate is generally 7 figure places, at least need to travel through 1,000 ten thousand car plates, calculated amount is large, cannot accomplish real-time, therefore use more limited.

Summary of the invention

The object of this invention is to provide a kind of similar vehicle license plate extraction method and device, by the similarity degree between kinds of characters, detect one group of license board information of specifying car plate similarity maximum with certain, effectively can reduce Car license recognition and artificially remember the identification error rate caused.

To achieve these goals, technical solution of the present invention is as follows:

A kind of similar vehicle license plate extraction method, for searching the car plate similar to it according to appointment car plate, comprises step:

Step 1, appointment car plate is split into character string dimension P _c, obtain character string dimension P _clength N, obtains character string dimension P _cthe first-phase of the similar character that each characters on license plate is corresponding and similar value composition thereof is like character matrix P _ewith the first similarity score matrix E _e;

Step 2, extract described first similarity score matrix E _ein the maximal value that often arranges, to sort from big to small composition array E by its value _es, according to array E _escalculate the most probable value that different number characters on license plate is erroneously identified, composition maximal value array A, and divide similar region successively according to the scope between maximal value array element A, each similar region is corresponding array A respectively ₁, A ₂..., A _n;

Step 3, according to described first-phase like character matrix P _ewith the first similarity score matrix E _e, calculate the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified successively, put into described array A respectively according to probable value size ₁, A ₂..., A _n, until array A ₁~ A _kmiddle number of elements is satisfied the demand the specified quantity of similar car plate obtained, and k is from 1 to N;

Step 4, according to described array A ₁~ A _kexport similar car plate.

Further, described step 3 comprises the steps:

Step 3.1, initialization iteration sequence number k are 1, start iteration;

The probable value that the various similar characters combinations that step 3.2, calculating k characters on license plate are erroneously identified are corresponding, puts into described array A respectively according to probable value size ₁, A ₂..., A _n;

Array A after step 3.3, calculating current iteration ₁~ A _kin element number num _kjudge num _kwhether be less than specified quantity, enter next step after if it is k being added 1, otherwise iteration terminate;

Step 3.4, judge whether k is greater than N, is, iteration terminates, otherwise return step 3.2.

Further, described step 3.2 specifically comprises step:

Step 3.2.1, for character string dimension P _cthe combinatorial matrix C of combination C (N, the k) correspondence of a middle k characters on license plate _k, combinatorial matrix C _keach row vector correspondence combination C (N, k) in one combination, suppose h correspondence combinatorial matrix C _kin the sequence number of the capable row vector of h, make h be 1 beginning iteration;

Step 3.2.2, according to the first similarity score matrix E _e, calculation combination Matrix C _kthe probable value of similar character combination corresponding to the capable row vector of h, form the second similarity score matrix R _k,h;

Step 3.2.3, by the second similarity score matrix R _k,hin probable value, put into array A respectively according to probable value size ₁, A ₂..., A _n;

Step 3.2.4, judge combinatorial matrix C _kin all combinations whether all traveled through, if traveled through, finishing iteration, otherwise return step 3.2.2 after h is added 1.

Further, for described second similarity score matrix R _k,h, suppose R _{k-1, h '}for the second similarity score matrix that k-1 characters on license plate in this combination is corresponding, described step 3.2.2 specifically comprises step:

According to kth-1 iteration result, find the second corresponding similarity score matrix R _{k-1, h '};

According to the first similarity score matrix E _eobtain the row R increased newly _en;

The second similarity score matrix R is obtained according to following formulae discovery _k,h:

R _k,h＝transform(R _k-1,h′·R _en ^T)

Wherein, 2≤k < N, R _enfor at the first similarity score matrix E _emiddle R _k,hrelative to R _{k-1, h '}newly-increased row, transform () is by multi-C vector of one-dimensional.Any second similarity score matrix R _k,hthe second similarity score matrix calculated in a front iteration can be utilized to calculate, decrease calculated amount.

Described according to described array A ₁~ A _kexport similar car plate, comprise step: to A ₁~ A _ksort respectively, get the element of described specified quantity according to probable value descending order, export the similar car plate of its correspondence.If need the specified quantity of the similar car plate of acquisition enough large, then, after the probable value all calculating the various similar character combination correspondences that 1 is erroneously identified to N number of characters on license plate terminates, export array A ₁, A ₂..., A _nin similar car plate corresponding to all element.Also can export the probable value of this similar car plate, i.e. similar car plate and the similar value of specifying car plate simultaneously.

The invention also proposes a kind of similar license plate retrieving device, for searching the car plate similar to it according to appointment car plate, this device comprises:

Car plate analysis module, for being split into character string dimension P by appointment car plate _c, obtain character string dimension P _clength N, obtains character string dimension P _cthe first-phase of the similar character that each characters on license plate is corresponding and similar value composition thereof is like character matrix P _ewith the first similarity score matrix E _e;

Division module, for extracting described first similarity score matrix E _ein the maximal value that often arranges, to sort from big to small composition array E by its value _es, according to array E _escalculate the most probable value that different number characters on license plate is erroneously identified, composition maximal value array A, and divide similar region successively according to the scope between maximal value array element A, each similar region is corresponding array A respectively ₁, A ₂..., A _n;

Computing module, for according to described first-phase like character matrix P _ewith the first similarity score matrix E _e, calculate the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified successively, put into described array A respectively according to probable value size ₁, A ₂..., A _n, until array A ₁~ A _kmiddle number of elements is satisfied the demand the specified quantity of similar car plate obtained, and k is from 1 to N.

Output module, for according to described array A ₁~ A _kexport similar car plate.

Further, described computing module, when the probable value of the various similar character combination correspondences that calculating 1 is erroneously identified to N number of characters on license plate, performs following operation steps:

Step I, initialization iteration sequence number k are 1, start iteration;

The probable value that the various similar characters combinations that step I i, calculating k characters on license plate are erroneously identified are corresponding, puts into described array A respectively according to probable value size ₁, A ₂..., A _n;

Array A after step I ii, calculating current iteration ₁~ A _kin element number num _kjudge num _kwhether be less than specified quantity, enter next step after if it is k being added 1, otherwise iteration terminate;

Step I v, judge whether k is greater than N, is, iteration terminates, otherwise return step I i.

Further, described computing module, when calculating the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified, performs following operation steps:

Step a, for character string dimension P _cthe combinatorial matrix C of combination C (N, the k) correspondence of a middle k characters on license plate _k, combinatorial matrix C _keach row vector correspondence combination C (N, k) in one combination, suppose h correspondence combinatorial matrix C _kin the sequence number of the capable row vector of h, make h be 1 beginning iteration;

Step b, according to the first similarity score matrix E _e, calculation combination Matrix C _kthe probable value of similar character combination corresponding to the capable row vector of h, form the second similarity score matrix R _k,h;

Step c, by the second similarity score matrix R _k,hin probable value, put into array A respectively according to probable value size ₁, A ₂..., A _n;

Steps d, judge combinatorial matrix C _kin all combinations whether all traveled through, if traveled through, finishing iteration, otherwise return step b after h is added 1.

Further, for described second similarity score matrix R _k,h, suppose R _{k-1, h '}for the second similarity score matrix that k-1 characters on license plate in this combination is corresponding, described computing module is at calculating second similarity score matrix R _k,htime, perform and operate as follows:

According to kth-1 iteration result, find the second corresponding similarity score matrix R _{k-1, h '};

According to the first similarity score matrix E _eobtain the row R increased newly _en;

The second similarity score matrix R is obtained according to following formulae discovery _k,h:

R _k,h＝transform(R _k-1,h′·R _en ^T)

Wherein, 2≤k < N, R _enfor at the first similarity score matrix E _emiddle R _k,hrelative to R _{k-1, h '}newly-increased row, transform () is by multi-C vector of one-dimensional.

Output module of the present invention specifically performs following operation: to A ₁~ A _ksort respectively, get the element of described specified quantity according to probable value descending order, export the similar car plate of its correspondence.If the specified quantity set is enough large, then, after the probable value all calculating the various similar characters combination correspondences that 1 is erroneously identified to N number of characters on license plate terminates, export array A ₁, A ₂..., A _nin similar car plate corresponding to all element, also can export its probable value, i.e. similar car plate and the similar value of specifying car plate simultaneously.

The similar vehicle license plate extraction method of one that the present invention proposes and device, based on LevenshteinDistance algorithm, similar car plate is calculated fast by multiple iteration, classification and ordination algorithm, greatly reduce Car license recognition, artificially remember reason, similar car plate exists the loss caused, the response time calculated fast is at about 1s.The present invention is by the similar matrix that gets actual under different condition and character matrix to adapt to different regions, and the similar car plate calculation requirement under different time sections, has stronger dirigibility and extendability.After iteration terminates, the array itself corresponding due to similar region is sequential, carries out sequence and is reduced to and sorts separately to certain several interval required, obtain the result after the sequence of similar car plate fast to all similar characters.

Accompanying drawing explanation

Fig. 1 is a kind of similar vehicle license plate extraction method process flow diagram of the present invention;

Fig. 2 is iterative algorithm process flow diagram of the present invention;

Fig. 3 is the present invention's specifically often kind of combined iteration calculation flow chart;

Fig. 4 is a kind of similar license plate retrieving apparatus structure schematic diagram of the present invention.

Embodiment

Be described in further details technical solution of the present invention below in conjunction with drawings and Examples, following examples do not form limitation of the invention.

In Car license recognition process, usually through the similarity degree between kinds of characters, similar character is found for uncertain character and replaces.The present invention, exactly by adding up a large amount of license plate recognition result, obtains the similarity score matrix of characters on license plate matrix and correspondence thereof, and is obtained the similar car plate of required specified quantity by iteration according to characters on license plate matrix and similarity score matrix.

As shown in Figure 1, a kind of similar vehicle license plate extraction method of the present embodiment comprises the steps:

Step 101, appointment car plate is split into character string dimension P _c, obtain character string dimension P _clength N, obtains character string dimension P _cthe first-phase of the similar character that each characters on license plate is corresponding and similar value composition thereof is like character matrix P _ewith the first similarity score matrix E _e.

Than if any car plate Zhejiang F12345 (following example is all for this car plate), split into character string dimension P _c=[Zhejiang F12345], wherein " Zhejiang F12345 " is characters on license plate, obtains character string dimension P _clength N=7.

By adding up a large amount of license plate recognition result, be not difficult to obtain for similar character corresponding to each characters on license plate and similar value thereof, by similar character matrix P corresponding for all characters on license plate of obtaining and similarity score matrix E:

$E=\left(\begin{array}{cccc}1& 1& ...& 1\\ e_{\mathrm{2,1}}& e_{\mathrm{2,2}}& ...& e_{2,n}\\ ...& ...& ...& ...\\ e_{m,1}& e_{m,2}& ...& e_{m,n}\end{array}\right)---\left(1\right)$

$P=\left(\begin{array}{cccc}{p}_{\mathrm{1,1}}& p_{\mathrm{1,2}}& p_{\mathrm{1,3}}& p_{1,n}\\ p_{\mathrm{2,1}}& p_{\mathrm{2,2}}& ...& p_{2,n}\\ ...& ...& ...& ...\\ p_{m,1}& p_{m,2}& ...& p_{m,n}\end{array}\right)---\left(2\right)$

P in P _{1, j}(j=1,2...n) is characters on license plate, p _i,j(i=2,3 ... m, j=1,2,3 ... n) be the similar character of characters on license plate, e in E _i,j(i=2,3 ... m; J=1,2,3 ... n) be similar character and the similar value of corresponding characters on license plate.The similar value that similar character is corresponding can be understood as the probable value that characters on license plate is identified as this similar character.Similar character matrix P and similarity score matrix form E, normally to the data that certain Recognition Algorithm of License Plate draws through statistical study the result of numerous Car license recognition, how similar character matrix P and similarity score matrix E is obtained, does not belong to emphasis of the present invention, do not repeat them here.

Now similarity score matrix E and characters on license plate matrix P is often arranged and sort from big to small by similar value, obtain the similarity score matrix E after orderly sequence _swith the similar character matrix P of correspondence _s:

E _s＝sort(E)(3)

P _s＝sort(P)(4)

Next by character string dimension P _ccorresponding first-phase is like character matrix P _ewith the first similarity score matrix E _esimilar character matrix P after sequence _swith the similarity score matrix E of correspondence _sin extract, and all filter out the first row, namely filter out similar value and the first row characters on license plate of characters on license plate itself respectively:

Such as formula (5) and formula (6) are exactly according to character string dimension P _cthe similarity score matrix E of=[Zhejiang F12345] after sequence _swith the similar character matrix P of correspondence _sin the first similarity score matrix E of extracting _ewith first-phase like character matrix P _e.E _ewith P _ehave position corresponding relation, wherein with " the Liao Dynasty ", 0.9 represents that the similar value of characters on license plate " Zhejiang " and " the Liao Dynasty " is 0.9, and the similar value of character " the Liao Dynasty " has showed the easy degree that character " Zhejiang " becomes " the Liao Dynasty ", and 0.9 is matrix E _ein maximum element, so character " Zhejiang " is known into " the Liao Dynasty " the most by mistake.

$\begin{array}{c}{E}_e=\left[\begin{array}{ccccccc}{e}_1& e_2& e_3& e_4& e_5& e_6& e_7\end{array}\right]\\ =\left[\begin{array}{ccccccc}0.9& 0.7& 0.8& 0.75& 0.65& 0.85& 0.6\\ 0.85& 0.65& 0.75& 0.7& 0.6& 0.8& 0.55\\ .& .& .& .& .& .& .\\ .& .& .& .& .& .& .\\ .& .& .& .& .& .& .\\ 0.3& 0.4& 0.35& 0.45& 0.4& 0.35& 0.3\end{array}\right]\end{array}---\left(5\right)$

It should be noted that and first similar character matrix P and similarity score matrix E is sorted, and then extract character string dimension P _ccorresponding first-phase is like character matrix P _ewith the first similarity score matrix E _e, because the first-phase extracted is like character matrix P _ewith the first similarity score matrix E _ebe and sorted, thus contribute to direct according to the first similarity score matrix E in subsequent steps _ethe first row extracts the maximum similar value often arranged.

The present embodiment extracts first-phase again like character matrix P after sorting to similar character matrix P and similarity score matrix E in addition _ewith the first similarity score matrix E _e, owing to not needing to sort after extraction at every turn, workload is little again.

Step 102, extract the first similarity score matrix E _ein the maximal value that often arranges, to sort from big to small composition array E by its value _es, according to array E _escalculate the most probable value that different number characters on license plate is erroneously identified, composition maximal value array A, and divide similar region successively according to the scope between maximal value array element A, each similar region is corresponding array A respectively ₁, A ₂..., A _n.

Because formula (5) is the result sorted to the similar value often arranged according to order from big to small, so formula (7) can be obtained from formula (5) matrix, namely get the first similarity score matrix E _ein the first row line ordering of going forward side by side obtain formula (7):

E _es＝[e _1maxe _2maxe _3maxe _4maxe _5maxe _6maxe _7max]

(7)

＝[0.90.850.80.750.70.650.6]

E _esin the most probable value that is erroneously identified of the characters on license plate of each its correspondence of element representation, such as 0.9 is the most probable value that characters on license plate " Zhejiang " is erroneously identified.

Obviously in order to obtain the maximum similar value of each row, except the first similarity score matrix E after above-mentioned acquisition sequence _emethod outside, also can not sort to similar character matrix P and similarity score matrix E, directly obtain the first similarity score matrix E _eafter get the maximum similar value that often arranges again to obtain array E _es.The invention is not restricted to adopt which kind of method, but extract the first similarity score matrix E again after similar character matrix P and similarity score matrix E is sorted _eobtain array E _e, workload is little.

It should be noted that the first similarity score matrix E here _ewhen inside often row order changes, first-phase is like character matrix P _ealso correspondence can change often row order, ensure position one_to_one corresponding (below roughly the same) all the time.

Obviously, according to array E _esknown, the most probable value only having a characters on license plate to be erroneously identified is 0.9, and the most probable value having two characters on license plate to be erroneously identified is 0.9*0.85, by that analogy, obtains maximal value array A:

$\begin{array}{c}A=\left[\begin{array}{ccccccc}{a}_1& a_2& a_3& a_4& a_5& a_6& a_7\end{array}\right]\\ =\left[\begin{array}{cccccc}0.9& 0.9*0.85& 0.9*0.85*0.8& 0.9*0.85*0.8*0.75& ...& \underset{i=1}{\overset{i=7}{\mathrm{\&Pi;}}}{e}_{i\max }\end{array}\right]\end{array}---\left(8\right)$

Wherein:

$\left\{\begin{array}{c}{a}_1=e_{1\max },e_{1\max }\&Element;E_{\mathrm{es}}\\ a_j=\underset{i=1}{\overset{i=j}{\mathrm{\&Pi;}}}{e}_{i\max },e_{i\max }\&Element;E_{\mathrm{es}},a_j\&Element;A,j=\mathrm{2,3},...,7\end{array}\right.---\left(9\right)$

Divide similar region successively according to the scope between maximal value array element A, the first similar region is [a ₁, a ₂], the second similar region is [a ₂, a ₃], by that analogy, comprise last similar region [a ₇, 0] and total N number of similar region.For existing 7 car plates of China, maximal value array A to there being 7 similar region, the corresponding array A of 7 similar region ₁, A ₂..., A ₇.

Step 103, according to described first-phase like character matrix P _ewith the first similarity score matrix E _e, calculate the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified successively, put into array A respectively according to probable value size ₁, A ₂..., A _n, until array A ₁~ A _kmiddle number of elements is satisfied the demand the specified quantity of similar car plate obtained, and k is from 1 to N.

The present embodiment is progressively found out and the number-plate number of specifying car plate similar by iteration successively, for the number-plate number of China 7 figure place, carry out at most 7 iteration, iteration goes out 1 ~ 7 character and is mistakenly identified as the situation of error character successively, until reach the specified quantity num needing the similar car plate obtained _resultor all iteration completes.

As shown in Figure 2, step 103 specifically comprises the steps:

Step 201, initialization iteration sequence number k are 1, start iteration.

Wherein k gets 1 to N, calculates respectively to the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified.

The probable value that the various similar characters combinations that step 202, calculating k characters on license plate are erroneously identified are corresponding, puts into array A respectively according to probable value size ₁, A ₂..., A _n.

Such as, when equaling 2 for k, 2 characters on license plate are erroneously identified, from 7 characters on license plate, select 2 characters on license plate, have in 21 and combine.Combine " Zhejiang F " for wherein a kind of, the similar character combination of its correspondence is the first similar matrix P _ethe combination in any of middle first row similar character and secondary series similar character, calculate 2 characters on license plate according to the first similarity score matrix and be erroneously identified the probable value that similar character combination is corresponding in situation, put into array A corresponding to each similar region of step 102 respectively according to probable value size ₁, A ₂..., A _n.

In detail see further describing below about step 202, as shown in Figure 3, specifically comprise the steps:

Step 301, for character string dimension P _cthe combinatorial matrix C of combination C (N, the k) correspondence of a middle k characters on license plate _k, combinatorial matrix C _keach row vector correspondence combination C (N, k) in one combination, suppose h correspondence combinatorial matrix C _kin the sequence number of the capable row vector of h, make h be 1 beginning iteration.

The character number of such as car plate Zhejiang F12345 is 7, and the combination of its different digit is respectively C (7,1), C (7,2) ..., C (7,7), such as C (7,2) just has Zhejiang A, Zhejiang F, Zhejiang 1, F1,45 grades are situation in 21 altogether.

When the character number combined is 3, one of them is combined as Zhejiang 23 to combine C (7,3), and it is at C ₃the row vector of middle correspondence is [1,0,0,1,1,0,0]; When the character number combined is 2, one of them is combined as Zhejiang 5 to combine C (7,2), and it is at C ₂the row vector of middle correspondence is [1,0,0,0,0,0,1]; C ₁be the unit matrix of 7 × 7, represent the combined situation of C (7,1).

The corresponding combinatorial matrix C of h _kin the sequence number of the capable row vector of h.

Step 302, according to the first similarity score matrix E _e, calculation combination Matrix C _kthe probable value of similar character combination corresponding to the capable row vector of h, form the second similarity score matrix R _k,h.

For C ₁, its first row vector is [1,0,0,0,0,0,0], and the second similarity score matrix of its correspondence is R ₁, ₁, be the first similarity score matrix E _efirst row R _e1:

$R_{e1}=\left[\begin{array}{c}0.9\\ 0.85\\ .\\ .\\ .\\ 0.3\end{array}\right]$

Correspondingly, for C ₁second row vector be [0,1,0,0,0,0,0], the second similarity score matrix of its correspondence is R _1,2, be the first similarity score matrix E _esecondary series R _e2, the second similarity score matrix that other row vectors are corresponding the like.

And for example for C ₂in combination " Zhejiang 5 " corresponding to the capable row vector of h, it is at C ₂the row vector of middle correspondence is [1,0,0,0,0,0,1], so can obtain the first similarity score matrix E _ein the 1st row R _e1with the 7th row R _e7, the second similarity score matrix R of the combination that the row vector that h is capable is corresponding _{2, h}for:

R _2,h＝transform(R _1,1·R _e7 ^T)

Wherein transform () is by multi-C vector of one-dimensional, such as:

$\left(\begin{array}{c}{R}_{e1}\\ R_{e2}\\ ...\\ R_{e7}\end{array}\right)=\mathrm{transform}\left((R_{e1},R_{e2},...,R_{e7})\right)$

For C ₂in the second similarity score matrix corresponding to other row vectors.

In like manner can obtain for C _ksecond similarity score matrix R of middle combination in any _k,h, wherein k is character number, k=1,2 ..., N, h are the one combination sequence number in C (N, k), and its value gets 1,2 ..., C (N, k).

It should be noted that, for any second similarity score matrix R of 2≤k < N _k,h, in second similarity score matrix that can calculate its front iteration (K-1 iteration), find a R _{k-1, h '}, and at the first similarity score matrix E _ein a newly-increased row R _en, and have:

R _k,h＝transform(R _k-1,h′·R _en ^T)

Such as, for combination C (7,3), one of them is combined as in " Zhejiang 23 ", and the row vector of its correspondence be [1,0,0,1,1,0,0], and it is relative to the combination " Zhejiang 2 " in combination C (7,2), increase newly at the first similarity score matrix E _ein be classified as R _e5, suppose that second similarity score matrix of " Zhejiang 2 " correspondence is R _{2, h '}, second similarity score matrix of " Zhejiang 23 " correspondence is R _{3, h}, then have: R _{3, h}=R _{2, h '}r _e5 ^t.Visible, any second similarity score matrix R _k,hthe second similarity score matrix calculated in a front iteration can be utilized to calculate, decrease calculated amount.

Step 303, by the second similarity score matrix R _k,hin probable value, put into array A respectively according to probable value size ₁, A ₂..., A _n.

Such as when first time iteration, k=1, the car plate that epicycle iterative computation only has a character similar, need the similar value traveling through all single characters, for similar character " Soviet Union " and " H ", their similar value is respectively 0.85 and 0.65, because the first similar region is [0.90.765], second similar region is [0.7650.612], because 0.85 belongs to the first similar region, then puts into array A by 0.85 ₁, because 0.65 belongs to the second similar region, then put into array A by 0.65 ₂.

When second time iteration, k=2, the car plate that epicycle iterative computation only has two characters similar, with combinatorial matrix C ₂in a combination " Zhejiang F " be example, its second similarity score matrix R _{2, h}for:

R after calculating the second similarity score matrix _{2, h}, respectively probable value is put into corresponding array, as 0.9 × 0.7 belongs to the second similar region, then puts into array A ₂.

Step 304, judge whether all combinations of epicycle iteration have all traveled through, if traveled through, finishing iteration, otherwise return step 302 after h is added 1.

Particularly, judging whether all combinations of epicycle iteration have all traveled through, is judge whether h is greater than this and takes turns C (N, k) combination sum corresponding to iteration, needs all combinations to travel through in an iteration.Such as C (7,2) is 21 combinations, then judge whether h is greater than 21, the like.

Array A after step 203, calculating current iteration ₁~ A _kin element number num _kjudge num _kwhether be less than specified quantity, enter next step after if it is k being added 1, otherwise iteration terminate.

After current kth time iteration completes, then judge array A ₁~ A _kin element number num _kwhether be less than specified quantity, if it is also want further iteration, otherwise just can according to array A ₁~ A _kin unit usually export the similar car plate of its correspondence.

After kth time iteration completes, A _kin element can not increase again, because probable value corresponding to the various similar characters combinations that are erroneously identified of k+1 character can not be greater than A _kthe probable value that middle element is corresponding, i.e. A ₁~ A _kelement in interval can not add again.

Step 204, judge whether k is greater than N, is, iteration terminates, otherwise return step 202.

If specified quantity is enough large, after the N time iteration, array A ₁, A ₂..., A _nin element number also do not reach specified quantity because characters on license plate only has N number of altogether, then no longer carry out iteration, iteration terminates.

It should be noted that, adopt above-mentioned iterative algorithm in the present embodiment to calculate the probable value of the 1 various similar character combination correspondences be erroneously identified to N number of characters on license plate, and put into corresponding array A respectively ₁, A ₂..., A _n.Due to phase array A ₁, A ₂..., A _nitself be sequential, sequence carried out to all similar characters and is reduced to required a few arrays are sorted separately, obtain the result after the sequence of similar car plate fast.But the present invention is not limited to above-mentioned iterative algorithm to carry out the calculating of probable value corresponding to various similar character combination, the situation that such as also can be erroneously identified k characters on license plate, calculate the probable value that often kind of similar character combination is corresponding one by one, without iterative computation, the invention is not restricted to whether use iterative algorithm or be not limited to concrete iterative algorithm, repeat no more here.

Step 104, according to array A ₁~ A _kexport similar car plate.

When having carried out k iteration, if array A ₁~ A _kin element number num _kequal the specified quantity needing the similar car plate obtained, that is just in time by array A ₁~ A _kin similar car plate corresponding to each element export.

If array A ₁~ A _kin element number num _kbe greater than the specified quantity of the similar car plate that needs obtain, then to A ₁~ A _ksort respectively, the similar car plate corresponding according to probable value descending order fetching determined number element exports.

If need the specified quantity of the similar car plate obtained enough large, then after N iteration completes by array A ₁, A ₂..., A _ncorresponding all similar car plates export.

It should be noted that, probable value corresponding to similar car plate can also be exported in the lump when exporting.Simultaneously due to array A ₁, A ₂..., A _nitself be (the first similar region > second similar region) that sort in order, so only need by the similar car plate of the Sequential output of array.

Fig. 4 shows the structural representation of a kind of similar license plate retrieving device of the present embodiment, and this device comprises:

Car plate analysis module, for being split into character string dimension P by appointment car plate _c, obtain character string dimension P _clength N, obtains character string dimension P _cthe first-phase of the similar character that each characters on license plate is corresponding and similar value composition thereof is like character matrix P _ewith the first similarity score matrix E _e;

Division module, for extracting the first similarity score matrix E _ein the maximal value that often arranges, to sort from big to small composition array E by its value _es, according to array E _escalculate the most probable value that different number characters on license plate is erroneously identified, composition maximal value array A, and divide similar region successively according to the scope between maximal value array element A, each similar region is corresponding array A respectively ₁, A ₂..., A _n;

Computing module, for according to first-phase like character matrix P _ewith the first similarity score matrix E _e, calculate the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified successively, put into described array A respectively according to probable value size ₁, A ₂..., A _n, until array A ₁~ A _kmiddle number of elements is satisfied the demand the specified quantity of similar car plate obtained, and k is from 1 to N;

Output module, for according to described array A ₁~ A _kexport similar car plate.

The present embodiment computing module, when calculating the probable value of the various similar character combination correspondences that calculating 1 is erroneously identified to N number of characters on license plate, performs all operations step of operation steps 103 in said method, repeats no more here.

Output module of the present invention specifically performs following operation:

To A ₁~ A _ksort respectively, get the element of described specified quantity according to probable value descending order, export the similar car plate of its correspondence.

Probable value corresponding to similar car plate can also be exported in the lump when exporting, if specified quantity is enough large, then after the probable value all calculating the various similar character combination correspondences that 1 is erroneously identified to N number of characters on license plate terminates, export the similar car plate of all probable values and correspondence thereof.

Above embodiment is only in order to illustrate technical scheme of the present invention but not to be limited; when not deviating from the present invention's spirit and essence thereof; those of ordinary skill in the art are when making various corresponding change and distortion according to the present invention, but these change accordingly and are out of shape the protection domain that all should belong to the claim appended by the present invention.

Claims (10)

1. a similar vehicle license plate extraction method, for searching the car plate similar to it according to appointment car plate, is characterized in that, comprise step:

Step 1, appointment car plate is split into character string dimension P _c, obtain character string dimension P _clength N, obtains character string dimension P _cthe first-phase of the similar character that each characters on license plate is corresponding and similar value composition thereof is like character matrix P _ewith the first similarity score matrix E _e;

Step 2, extract described first similarity score matrix E _ein the maximal value that often arranges, to sort from big to small composition array E by its value _es, according to array E _escalculate the most probable value that different number characters on license plate is erroneously identified, composition maximal value array A, and divide similar region successively according to the scope between maximal value array element A, each similar region is corresponding array A respectively ₁, A ₂..., A _n;

Step 3, according to described first-phase like character matrix P _ewith the first similarity score matrix E _e, calculate the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified successively, put into described array A respectively according to probable value size ₁, A ₂..., A _n, until array A ₁~ A _kmiddle number of elements is satisfied the demand the specified quantity of similar car plate obtained, and k is from 1 to N;

Step 4, according to described array A ₁~ A _kexport similar car plate.

2. similar vehicle license plate extraction method according to claim 1, is characterized in that, described step 3 comprises the steps:

Step 3.1, initialization iteration sequence number k are 1, start iteration;

The probable value that the various similar characters combinations that step 3.2, calculating k characters on license plate are erroneously identified are corresponding, puts into described array A respectively according to probable value size ₁, A ₂..., A _n;

Array A after step 3.3, calculating current iteration ₁~ A _kin element number num _kjudge num _kwhether be less than specified quantity, enter next step after if it is k being added 1, otherwise iteration terminate;

Step 3.4, judge whether k is greater than N, is, iteration terminates, otherwise return step 3.2.

3. similar vehicle license plate extraction method according to claim 2, is characterized in that, described step 3.2 specifically comprises step:

Step 3.2.1, for character string dimension P _cthe combinatorial matrix C of combination C (N, the k) correspondence of a middle k characters on license plate _k, combinatorial matrix C _keach row vector correspondence combination C (N, k) in one combination, suppose h correspondence combinatorial matrix C _kin the sequence number of the capable row vector of h, make h be 1 beginning iteration;

Step 3.2.2, according to the first similarity score matrix E _e, calculation combination Matrix C _kthe probable value of similar character combination corresponding to the capable row vector of h, form the second similarity score matrix R _k,h;

Step 3.2.3, by the second similarity score matrix R _k,hin probable value, put into array A respectively according to probable value size ₁, A ₂..., A _n;

Step 3.2.4, judge combinatorial matrix C _kin all combinations whether all traveled through, if traveled through, finishing iteration, otherwise return step 3.2.2 after h is added 1.

4. similar vehicle license plate extraction method according to claim 3, is characterized in that, for described second similarity score matrix R _k,h, suppose R _{k-1, h '}for the second similarity score matrix that k-1 characters on license plate in this combination is corresponding, described step 3.2.2 specifically comprises step:

According to kth-1 iteration result, find the second corresponding similarity score matrix R _{k-1, h '};

According to the first similarity score matrix E _eobtain the row R increased newly _en;

The second similarity score matrix R is obtained according to following formulae discovery _k,h:

R _k,h＝transform(R _k-1,h′·R _en ^T)

Wherein, 2≤k < N, R _enfor at the first similarity score matrix E _emiddle R _k,hrelative to R _{k-1, h '}newly-increased row, transform () is by multi-C vector of one-dimensional.

5. similar vehicle license plate extraction method according to claim 1, is characterized in that, described according to described array A ₁~ A _kexport similar car plate, comprise step:

To A ₁~ A _ksort respectively, get the element of described specified quantity according to probable value descending order, export the similar car plate of its correspondence.

6. a similar license plate retrieving device, for searching the car plate similar to it according to appointment car plate, it is characterized in that, this device comprises:

Car plate analysis module, for being split into character string dimension P by appointment car plate _c, obtain character string dimension P _clength N, obtains character string dimension P _cthe first-phase of the similar character that each characters on license plate is corresponding and similar value composition thereof is like character matrix P _ewith the first similarity score matrix E _e;

Division module, for extracting described first similarity score matrix E _ein the maximal value that often arranges, to sort from big to small composition array E by its value _es, according to array E _escalculate the most probable value that different number characters on license plate is erroneously identified, composition maximal value array A, and divide similar region successively according to the scope between maximal value array element A, each similar region is corresponding array A respectively ₁, A ₂..., A _n;

Computing module, for according to described first-phase like character matrix P _ewith the first similarity score matrix E _e, calculate the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified successively, put into described array A respectively according to probable value size ₁, A ₂..., A _n, until array A ₁~ A _kmiddle number of elements is satisfied the demand the specified quantity of similar car plate obtained, and k is from 1 to N;

Output module, for according to described array A ₁~ A _kexport similar car plate.

7. similar license plate retrieving device according to claim 6, is characterized in that, described computing module, when the probable value of the various similar character combination correspondences that calculating 1 is erroneously identified to N number of characters on license plate, performs following operation steps:

Step I, initialization iteration sequence number k are 1, start iteration;

The probable value that the various similar characters combinations that step I i, calculating k characters on license plate are erroneously identified are corresponding, puts into described array A respectively according to probable value size ₁, A ₂..., A _n;

Array A after step I ii, calculating current iteration ₁~ A _kin element number num _kjudge num _kwhether be less than specified quantity, enter next step after if it is k being added 1, otherwise iteration terminate;

Step I v, judge whether k is greater than N, is, iteration terminates, otherwise return step I i.

8. similar license plate retrieving device according to claim 7, is characterized in that, described computing module, when calculating the probable value of the various similar character combination correspondences that k characters on license plate is erroneously identified, performs following operation steps:

Step a, for character string dimension P _cthe combinatorial matrix C of combination C (N, the k) correspondence of a middle k characters on license plate _k, combinatorial matrix C _keach row vector correspondence combination C (N, k) in one combination, suppose h correspondence combinatorial matrix C _kin the sequence number of the capable row vector of h, make h be 1 beginning iteration;

Step b, according to the first similarity score matrix E _e, calculation combination Matrix C _kthe probable value of similar character combination corresponding to the capable row vector of h, form the second similarity score matrix R _k,h;

Step c, by the second similarity score matrix R _k,hin probable value, put into array A respectively according to probable value size ₁, A ₂..., A _n;

Steps d, judge combinatorial matrix C _kin all combinations whether all traveled through, if traveled through, finishing iteration, otherwise return step b after h is added 1.

9. similar license plate retrieving device according to claim 8, is characterized in that, for described second similarity score matrix R _k,h, suppose R _{k-1, h '}for the second similarity score matrix that k-1 characters on license plate in this combination is corresponding, described computing module is at calculating second similarity score matrix R _k,htime, perform and operate as follows:

According to kth-1 iteration result, find the second corresponding similarity score matrix R _{k-1, h '};

According to the first similarity score matrix E _eobtain the row R increased newly _en;

The second similarity score matrix R is obtained according to following formulae discovery _k,h:

R _k,h＝transform(R _k-1,h′·R _en ^T)

Wherein, 2≤k < N, R _enfor at the first similarity score matrix E _emiddle R _k,hrelative to R _{k-1, h '}newly-increased row, transform () is by multi-C vector of one-dimensional.

10. similar license plate retrieving device according to claim 6, is characterized in that, described output module performs following operation:

To A ₁~ A _ksort respectively, get the element of described specified quantity according to probable value descending order, export the similar car plate of its correspondence.