CN107358231A - A kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm - Google Patents
A kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm Download PDFInfo
- Publication number
- CN107358231A CN107358231A CN201710429590.3A CN201710429590A CN107358231A CN 107358231 A CN107358231 A CN 107358231A CN 201710429590 A CN201710429590 A CN 201710429590A CN 107358231 A CN107358231 A CN 107358231A
- Authority
- CN
- China
- Prior art keywords
- mrow
- msup
- chaos
- sift
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
Abstract
The invention belongs to Vehicle License Plate Recognition System, is a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm, is characterized in:The Chinese character of car plate and digital alphabet are separately identified, i.e., Chinese character identified using the feature extraction of SIFT operators and template matching method;Digital alphabet is identified using 13 point feature extraction methods and SVMs, the problem of major part is identified using unified character feature extraction and recognition methods, causes the overall discrimination of characters on license plate low, is solved in the Vehicle License Plate Recognition System of prior art.Simultaneously, to improve the classification capacity of SVMs, its Radial basis kernel function parameter and penalty factor are optimized using Chaos Genetic Algorithm, gather the license plate image under different background, test simulation is carried out on matlab softwares, more than 99% can be reached by drawing the overall discrimination of character, and Chaos Genetic Algorithm is higher than the character identification rate of traditional genetic algorithm, fast convergence rate.
Description
Technical field
The invention belongs to Vehicle License Plate Recognition System, is a kind of Recognition of License Plate Characters based on SIFT operators and Chaos Genetic Algorithm
Method.
Background technology
Vehicle License Plate Recognition System is capturing violating the regulations, charge, the monitoring of parking lot gateway, car speed and flow control etc. at a high speed
Various occasions have turned into the Main Means of intelligent traffic administration system.Automobile industry flourishes under the drive of national economy, vehicle
Quantity also increases sharply, but brings the traffic problems of getting worse, such as drives over the speed limit, parking offense, makes a dash across the red light, traffic thing
Therefore, charge confusion etc., therefore, Vehicle License Plate Recognition System is also required to constantly innovation and improved.2012, Guo Jinzhi was in publication source
The distribution unit of Xian Electronics Science and Technology University proposes the Vehicle License Plate Recognition System based on SIFT algorithms, and the character of car plate is uniformly adopted
Be identified with the mode of SIFT operator extractions feature and template matches, due to Chinese character and digital alphabet structure complexity and
Local feature richness difference causes both discriminations to truly have obvious difference, and result is Chinese Character Recognition rate height, digital alphabet
Discrimination is substantially low, and then overall discrimination is not high.
The content of the invention
The technical problems to be solved by the invention are:The shortcomings that overcoming prior art, there is provided one kind based on SIFT operators and
The license plate character recognition method of Chaos Genetic Algorithm, Chinese character and digital alphabet can separately be identified, it is overall to improve characters on license plate
Discrimination.
The present invention solve technical problem scheme be:It is a kind of to be known based on SIFT operators and the characters on license plate of Chaos Genetic Algorithm
Other method, including Chinese Character Recognition, the Chinese Character Recognition include two steps:Extract the pass that SIFT operators feature forms character feature
Key point;The key point of the character feature of formation and the template matches of car plate Chinese character are subjected to Chinese Character Recognition, it is characterized in that:Also include
Digital alphabet identifies, comprises the following steps that:
1) SIFT operator features are extracted, form the key point of character feature;
2) key point of the character feature of formation and the template matches of car plate Chinese character are subjected to Chinese Character Recognition;
3) alphanumeric characters are extracted;
4) identification of alphanumeric characters;
5) Chinese Character Recognition and digital Letter identification combine to form number-plate number identification.
The step of step 1) extraction SIFT operator features, is as follows:
I structure metric space:Assuming that I (x, y) is input picture, G (x, y, σ) is the Gaussian function of mutative scale, then the figure
The metric space function L (x, y, σ) of picture definition is following formula:
L (x, y, σ)=G (x, y, σ) * I (x, y) (1)
In formula, σ is scale factor, and * is convolution algorithm, and (x, y) represents location of pixels coordinate in image;
Gaussian function G (x, y, σ) is defined as following formula:
In order to effectively detect stable extremal point in metric space, using difference of Gaussian function D (x, y, σ), it is logical
Cross what the image subtraction of two adjacent k times of yardsticks differences was obtained:
D (x, y, σ)=(G (x, y, k σ)-G (x, y, σ)) * I (x, y)=L (x, y, k σ)-L (x, y, σ) (3)
II Local Extremum detects:In order to detect the local maximum of difference of Gaussian function and minimum, each sampled point
Will be compared with it be with 8 consecutive points of yardstick and 9 × 2, totally 26 points of neighbouring yardstick, when the sampled point
It is all higher than or when less than other, the sampled point is selected, therefore can detect local pole all in the space
It is worth point, as SIFT feature candidate point;
III is accurately positioned extreme point:The candidate's SIFT feature detected to previous step is screened, and is filtered out to noise
The point lower than more sensitive, contrast and unstable marginal point, further to obtain the accurate point in position;
The screening technique of candidate's SIFT feature is to be fitted DOG spaces local sampling, inspection using three-dimensional quadratic equation
Survey and remove low contrast point, carrying out Taylor expansion to quadratic term to D (x, y, σ) obtains:
To above formula derivation, and make equation be equal to zero, D (x) extreme point can be obtainedIt is shown below:
Formula (5) is substituted into formula (4) and obtains formula (6):
If it is calculatedThen this feature point remains, and otherwise removes, and stays pinpoint pole
Value point is exactly key point;
IV key point direction is distributed and generation SIFT feature vector
, it is necessary to determine one of key point according to the size of its neighborhood territory pixel and directional spreding situation after key point determination
Principal direction, to realize the rotational invariance of SIFT operators;
Any one key point L (x, y) can calculate its gradient modulus value m (x, y) and direction θ (x, y), see formula
And (8) (7):
θ (x, y)=tan-1(L(x,y+1)-L(x.y-1))/(L(x+1,y)-L(x-1,y)) (8)
The gradient modulus value m (x, y) and direction θ (x, y) that are calculated according to formula (7) and (8) are the principal direction of key point.
, will when the key point for the character feature that the step 2) is formed and the template matches of car plate Chinese character carry out Chinese Character Recognition
Data after the feature extraction of each character are imported matlab and matched using nearest neighbor algorithm, when both matching numbers most
When more, corresponding car plate Chinese character is character to be identified.
Step 3) the extraction alphanumeric characters use ten line-of-sight courses.
Step 4) the alphanumeric characters are identified using SVMs, are comprised the following steps:
(1) formula is seen below using Radial basis kernel function generation SVMs, Radial basis kernel function:
K(x,xi)=exp-| | x-xi||2/2σ2} (9)
In formula:Exp represents the exponential function using natural constant e the bottom of as;X represents that pixel is thinking direction coordinate;xiRepresent
Coordinate of the ith pixel point in x directions;σ represents nuclear parameter.
(2) optimized with Chaos Genetic Algorithm and determine penalty factor and nuclear parameter σ;
(3) vector machine classification design is supported, to determine three class graders;
(4) the three class graders that the penalty factor of step (2) optimization and nuclear parameter σ are imported to SVMs are known
Not.
The step (2) optimizes penalty factor with Chaos Genetic Algorithm and nuclear parameter σ comprises the following steps:
1. determine fitness function
Using the Radial basis kernel function of formula (9) as fitness function;
2. Initial parameter sets
Determine operational factor and population scale M;
Using binary coding, it is integer, σ ∈ [0.0001,1], penalty factor and nuclear parameter σ to select C ∈ [1,100]
Parameter group represent that the population size of its initial population is 80 with 21 bits;
Maximum iteration is n=100;
Crossover probability PcFor 0.8;
Mutation probability
Wherein L is chromosome length;
3. Logistic chaotic maps produce initial population
Logistic chaotic maps, its generation initial population mapping equation are:
xn+1=μ xn(1-xn) n=0,1,2 ... (11)
In formula, xnThe relative number of n rear individual of iteration is represented, μ is controling parameter, is takenWhen, Equation Iterative fortune
Dynamic rail mark will be in Complete Chaos motion state;
4. calculate individual adaptation degree using algorithm of support vector machine
Individual adaptation degree is defined as formula:
5. crossover operation is carried out to population
By individual random combine two-by-two in population, the combination to each pairing, generated at random by system first one (0,1)
Between number, decide whether to intersect by crossover probability.If intersecting, height is utilized after simple mapping using the sequence of mapping generation
This function determines crossover location, and otherwise, lower a pair of the combinations of matching, all crossover locations can be determined by a chaos sequence
It is fixed;
6. mutation operator is carried out to current population
Decide whether to morph with mutation probability by the number that generates at random first, mapped to Logistic and assign n
The initial value of fine difference, n chaos sequence will be generated;
7. meet that stop condition then obtains the SVMs penalty factor of optimization and nuclear parameter σ combination parameter;
8. decode the penalty factor and nuclear parameter σ optimized;
9. if stop condition is unsatisfactory for for 4. individual adaptation degree that step calculates, it is necessary to reselect and calculate.
4. individual adaptation degree that the step calculates is unsatisfactory for stop condition, needs to reselect and calculate, its step
Including:
A chaos intialization populations
An initial vector, X are randomly generated firsti=(xi1,xi2,…xim), by XiObtained according to Logistic mapping equations
To random initializtion population:
By N number of vectorial X1,X2,…XNThe span section of objective optimization function requirements initializaing variable is transformed to, obtains N
Individual initializaing variable is designated as population at individual quantity as initial population;
B chaotic crossovers
Control the frequency of crossover operation with chaos sequence, concrete operations are as follows:Using chaotic crossover mapping equation:
xn+1=4xn(1-xn) (14)
Appoint and take an initial value x0, it can produce a sequence of iterations, the characteristics of this sequence be changed between [0,1],
Obviously, can be by xn+1As a random switching, work as xn+1During more than the value selected in advance, with regard to carrying out crossover operation;Conversely,
Do not intersect then, i.e.,:
P is previously selected value in above formula, typically takes 0.5;
C chaotic mutations
In simple hereditary calculation method, mutation probability pmSelection be random, or by caused chaos sequence Lai
Control the progress of mutation operation;
D replicates excellent individual in parent colony and new colony and enters the next generation.
Step (3) the support vector cassification design is using one-to-one integrated mode, it is necessary to three class graders of construction
Quantity is:
In formula:N is natural number, value 10.
Beneficial effects of the present invention are embodied in the following aspects:
1 present invention separates the Chinese character and digital alphabet of car plate, SIHT operator identifications Chinese character is respectively adopted, using ten three points
Feature extraction and SVMs identification digital alphabet, recombinant form the number-plate number and know method for distinguishing, solve existing skill
It is most of in the Vehicle License Plate Recognition System of art to be identified using unified character feature extraction and recognition methods, cause car plate word
The problem of overall discrimination is low is accorded with,;
2 the present invention using SIFT algorithms extraction carry out car plate Chinese character feature extractions, and by the key point of character feature with
The template matches of car plate Chinese character are matched using nearest neighbor algorithm, so as to carry out Chinese Character Recognition, are taken full advantage of Chinese character and are tied in itself
Structure is complicated, local feature information clearly and it is abundant the characteristics of, while had the advantage that using SIFT operator feature extractions, it is all
Yardstick and rotation such as image maintain the invariance, the object features few to quantity extraction can also obtain feature more than comparison
Point, characteristic point information are abundant, unique good, accurately quickly can be matched in substantial amounts of data, so as to realize the car plate Chinese
The quick identification of word, improves the discrimination of car plate Chinese character, and discrimination can reach 100%.
3 present invention are extracted with 13 point feature extraction methods to alphanumeric characters, and digital alphabet is carried out with SVMs
Feature recognition, can according to digital alphabet it is simple in construction the characteristics of, simplify feature extraction difficulty, and according to Vnpnik structure risks
Minimization principle is classified, and small error can be obtained by limited training sample.It is asked using training error as optimization
The constraints of topic, minimized using fiducial range value as optimization aim, its solution, which can regard solution one as, to be had linearly about
The problem of quadratic convex programming of beam, and local minimum is not present, can be SVMs by introducing different kernel functions
Linear solution be transformed into non-linear solution, therefore high-dimensional sample is calculated will not increase amount of calculation, with Chaos-Genetic
The penalty factor and nuclear parameter σ of algorithm optimization, fast convergence rate, improve character identification rate;
The penalty factor and nuclear parameter σ of 4 SVMs are the key parameters for influenceing support vector cassification performance, are changed
Becoming the value of penalty factor and nuclear parameter σ can make character identification rate change, and the two values how be chosen, so that discrimination
It is an optimization problem to reach maximum.The present invention is utilized simultaneously using Chaos Genetic Algorithm optimization penalty factor and nuclear parameter σ
Genetic algorithm with concurrency, robustness, uncertainty, integrity and optimization the advantages of and chaos with it is sensitive to initial value
The characteristic such as property, ergodic, randomness, by incorporating chaos algorithm, the sensitiveness with chaos algorithm to initial value, makes it produce
The initial population of even distribution solution space, can avoid optimization calculating from being absorbed in local optimum, and then effectively solve genetic algorithm
Precocious phenomenon;With the randomness of chaos algorithm, it is set to produce initial population used in the genetic algorithm of completely random;Chaos
The ergodic of algorithm, makes the every nook and cranny of its caused population traversal solution space, and utilizes this characteristic pair of chaos algorithm
Newborn population is searched for completely, algorithm is able to maintain that the diversity of population when carrying out genetic manipulation, beneficial to raising algorithm
Convergence rate, it is random to overcome initial population existing for genetic algorithm, make it big in operation early stage individual difference and
Its usual selection operation mode is roulette mode, can cause precocious phenomenon, tend to one in Evolution of Population later stage, fitness
Cause, excellent individual offspring advantage unobvious in evolution, cause the slow-footed defect of whole convergence in population, make penalty factor and
Nuclear parameter σ is optimal, so as to improve the discrimination of digital alphabet.
4. individual adaptation degree that the step of 5 present invention calculates is unsatisfactory for stop condition, needs what is reselected and calculate
Process, it can be reselected most by chaos intialization population quantity, chaos optimization crossover operation, chaos optimization mutation operation
Excellent individual adaptation degree;
The support vector cassification design of 6 present invention uses one-to-one integrated mode, and it is more to construct suitable supporting vector
Grader, meet that Recognition of License Plate Characters belongs to the needs of more classification problems.
Brief description of the drawings
Fig. 1 is the license plate character recognition method flow chart of the present invention;
Fig. 2 is the present invention using Chaos Genetic Algorithm optimization penalty factor and nuclear parameter σ flow chart;
The Local Extremum detects schematic diagram of Fig. 3 embodiments 1;
The SIHT feature extraction figures of the part Chinese character of Fig. 4 embodiments 1;
The SIFT feature template matches figure of Fig. 5 embodiments 1 " capital " word;
The fitness curve of 1 two kinds of algorithms of Fig. 6 embodiments;
Fig. 7 embodiment 1GA-SVM algorithm numeral recognition result figures;
Fig. 8 embodiment 1CGA-SVM algorithm numeral recognition result figures;
The SIFT feature template matches figure of Fig. 9 embodiments 2 " Ji " word.
Embodiment
With reference to embodiment, the present invention is further described.
Referring to Fig. 1-Fig. 8, embodiment 1, a kind of characters on license plate based on SIFT operators and Chaos Genetic Algorithm of the present embodiment
Recognition methods, the SIFT feature template matches figure of the Chinese character part for selecting Jilin, as shown in figure 4, its method includes the Chinese
Word identifies and digital Letter identification, initially sets up the car plate Chinese character ATL of standard, China each province and city, autonomous region totally three ten one
Chinese character, by putting into practice collection, clearly car plate Chinese character picture is used as car plate Chinese character STL, and picture size is unified for 42 ×
21 pixels, are comprised the following steps that:
1) SIFT operator features are extracted, form the key point of character feature;
The extraction SIFT operator features follow these steps to carry out:
I structure metric space:Assuming that I (x, y) is input picture, G (x, y, σ) is the Gaussian function of mutative scale, then the figure
The metric space function L (x, y, σ) of picture definition is following formula:
L (x, y, σ)=G (x, y, σ) * I (x, y) (1)
In formula, σ is scale factor, and * is convolution algorithm, and (x, y) represents location of pixels coordinate in image;
Gaussian function G (x, y, σ) is defined as following formula:
In order to effectively detect stable extremal point in metric space, using difference of Gaussian function D (x, y, σ), it is logical
Cross what the image subtraction of two adjacent k times of yardsticks differences was obtained:
D (x, y, σ)=(G (x, y, k σ)-G (x, y, σ)) * I (x, y)=L (x, y, k σ)-L (x, y, σ) (3)
II Local Extremum detects:In order to detect the local maximum of difference of Gaussian function and minimum, each sampled point
Will be compared with it be with 8 consecutive points of yardstick and 9 × 2, totally 26 points of neighbouring yardstick, when the sampled point
It is all higher than or when less than other, the sampled point is selected, therefore can detect local pole all in the space
It is worth point, as SIFT feature candidate point, as shown in Figure 3;
III is accurately positioned extreme point:The candidate's SIFT feature detected to previous step is screened, and is filtered out to noise
The point lower than more sensitive, contrast and unstable marginal point, further to obtain the accurate point in position;
The screening technique of candidate's SIFT feature is to be fitted DOG spaces local sampling, inspection using three-dimensional quadratic equation
Survey and remove low contrast point, carrying out Taylor expansion to quadratic term to D (x, y, σ) obtains:
To above formula derivation, and make equation be equal to zero, D (x) extreme point can be obtainedIt is shown below:
Formula (5) is substituted into formula (4) and obtains formula (6):
If it is calculatedThen this feature point remains, and otherwise removes, and stays pinpoint pole
Value point is exactly key point;
IV key point direction is distributed and generation SIFT feature vector
, it is necessary to determine one of key point according to the size of its neighborhood territory pixel and directional spreding situation after key point determination
Principal direction, to realize the rotational invariance of SIFT operators;
Any one key point L (x, y) can calculate its gradient modulus value m (x, y) and direction θ (x, y), see formula
And (8) (7):
θ (x, y)=tan-1(L(x,y+1)-L(x.y-1))/(L(x+1,y)-L(x-1,y)) (8)
The gradient modulus value m (x, y) and direction θ (x, y) that are calculated according to formula (7) and (8) are the principal direction of key point.
2) key point of the character feature of formation and the template matches of car plate Chinese character are subjected to Chinese Character Recognition;
When the key point of the character feature of the formation and the template matches of car plate Chinese character carry out Chinese Character Recognition, by each character
Feature extraction after data import matlab and matched using nearest neighbor algorithm, when both matching numbers are most, institute
Corresponding car plate Chinese character is character to be identified.
3) alphanumeric characters are extracted;
Extraction alphanumeric characters use ten line-of-sight courses;
4) identification of alphanumeric characters;
The alphanumeric characters use the identification that SVMs is carried out, and comprise the following steps:
(1) formula is seen below using Radial basis kernel function generation SVMs, Radial basis kernel function:
K(x,xi)=exp-| | x-xi||2/2σ2}(9)
In formula:Exp represents the exponential function using natural constant e the bottom of as;X represents that pixel is thinking direction coordinate;xiRepresent
Coordinate of the ith pixel point in x directions;σ represents nuclear parameter.
(2) optimized with Chaos Genetic Algorithm and determine penalty factor and nuclear parameter σ;
The Chaos Genetic Algorithm optimization penalty factor and nuclear parameter σ comprise the following steps:
1. determine fitness function
Using the Radial basis kernel function of formula (9) as fitness function;
2. Initial parameter sets
Determine operational factor and population scale M;
Using binary coding, it is integer, σ ∈ [0.0001,1], penalty factor and nuclear parameter σ to select C ∈ [1,100]
Parameter group represent that the population size of its initial population is 80 with 21 bits;
Maximum iteration is n=100;
Crossover probability PcFor 0.8;
Mutation probability
Wherein L is chromosome length;
3. Logistic chaotic maps produce initial population
Logistic chaotic maps, its generation initial population mapping equation are:
xn+1=μ xn(1-xn) n=0,1,2 ... (11)
In formula, xnThe relative number of n rear individual of iteration is represented, μ is controling parameter, is takenWhen, Equation Iterative fortune
Dynamic rail mark will be in Complete Chaos motion state;
4. calculate individual adaptation degree using algorithm of support vector machine
Individual adaptation degree is defined as formula:
5. crossover operation is carried out to population
By individual random combine two-by-two in population, the combination to each pairing, generated at random by system first one (0,1)
Between number, decide whether to intersect by crossover probability.If intersecting, height is utilized after simple mapping using the sequence of mapping generation
This function determines crossover location, and otherwise, lower a pair of the combinations of matching, all crossover locations can be determined by a chaos sequence
It is fixed;
6. mutation operator is carried out to current population
Decide whether to morph with mutation probability by the number that generates at random first, mapped to Logistic and assign n
The initial value of fine difference, n chaos sequence will be generated;
7. meet that stop condition then obtains the SVMs penalty factor of optimization and nuclear parameter σ combination parameter;
8. decode the penalty factor and nuclear parameter σ optimized;
9. if stop condition is unsatisfactory for for 4. individual adaptation degree that step calculates, it is necessary to reselect and calculate.
4. individual adaptation degree that the step calculates is unsatisfactory for stop condition, needs to reselect and calculate, its step
Including:
A chaos intialization populations
An initial vector, X are randomly generated firsti=(xi1,xi2,…xim), by XiObtained according to Logistic mapping equations
To random initializtion population:
By N number of vectorial X1,X2,…XNThe span section of objective optimization function requirements initializaing variable is transformed to, obtains N
Individual initializaing variable is designated as population at individual quantity as initial population;
B chaotic crossovers
Control the frequency of crossover operation with chaos sequence, concrete operations are as follows:Using chaotic crossover mapping equation:
xn+1=4xn(1-xn) (14)
Appoint and take an initial value x0, it can produce a sequence of iterations, the characteristics of this sequence be changed between [0,1],
Obviously, can be by xn+1As a random switching, work as xn+1During more than the value selected in advance, with regard to carrying out crossover operation;Conversely,
Do not intersect then, i.e.,:
P is previously selected value in above formula, typically takes 0.5;
C chaotic mutations
In simple hereditary calculation method, mutation probability pmSelection be random, or by caused chaos sequence Lai
Control the progress of mutation operation;
D replicates excellent individual in parent colony and new colony and enters the next generation..
(3) vector machine classification design is supported, to determine three class graders;
Step (3) support vector cassification design uses one-to-one integrated mode, car plate totally 0~90 numerals, needs
The three class grader quantity to be constructed are:
In formula:N=10.
(4) the three class graders that the penalty factor of step (2) optimization and nuclear parameter σ are imported to SVMs are known
Not.
5) Chinese Character Recognition and digital Letter identification combine to form number-plate number identification.
It is to be identified by being classified as with the most template of car plate Chinese character number of matches when the present embodiment carries out Chinese Character Recognition
Chinese character classification, the matching points of remaining Chinese character are obvious less, match point entanglement or concentrate matching.In order to verify that this method Chinese character is known
Other accuracy rate, the car plate picture that the present embodiment is gathered under different scenes are tested such as table 1, it can be seen that SIFT Hanzi featureses
The mode of extraction and template matches is especially suitable for the identification of car plate Chinese character, and discrimination is approximately 100%.
Chinese Character Recognition rate under the different background of table 1
When the present embodiment carries out digital alphabet identification, using the LIBSVM software development kits of professor's Lin Zhiren exploitation to character
Sample data is trained.
The experimental situation of the present embodiment:Dominant frequency 2.2GHz, internal memory 2G computer, MATLAB R2014a platforms.This implementation
The experimental data of example comes from captured car plate numerical character, and 20 classes are chosen from digital alphabet, per class 50 groups of samples of character, wherein
20 groups of data are used to train, and 30 groups of data are used to test in addition.Two kinds of algorithm routines of GA-SVM and CGA-SVM are utilized respectively to carry out
Training.In training process, optimized algorithm can continuous transformation parameter value, fitness is reached maximum.Calculated to find
Above two algorithm is respectively trained 20 times for the regularity of method optimization, the present embodiment, takes out the result of maximum probability as analysis
Object, Fig. 6 give the fitness of two kinds of algorithms with the change curve of evolutionary generation.Embodiment be based on identical population quantity,
On the premise of evolutionary generation and experimental data, GA reaches first locally optimal solution in 7 generation particles, afterwards in the 16th generation, 22 generations
Local optimum is jumped out Deng multiple, but final fitness value is less than CGA.CGA convergence rates are very fast, just converged in the 58th generation particle
Highest fitness.Although CGA enters local optimum in the 27th generation and 57 generations, CGA can be jumped out successfully.And whole
In convergence process, GA fitness curves are more tortuous, and by contrast, CGA fitness curve ladders are less, illustrate in searching process
Middle CGA is not easy to be absorbed in local optimum.So in SVM training process, CGA just reaches global optimum in the 58th generation, with GA
Algorithm is compared, and CGA is attained by ideal degree in terms of convergence rate and parameter optimal value.
SVMs after two kinds of algorithm optimizations of GA and CGA is subjected to character recognition to test data, respectively to numeral
Totally 10 characters, totally 34 characters establish respective grader to letter (remove I and O) respectively for totally 24 characters, digital alphabets, per class
Character totally 60 samples, first 30 are used to train, and latter 30 are used to test.Classification and Identification result such as Fig. 7, Fig. 8 and table 2.
The classification results statistical form of the two methods of table 2
After intuitively reflecting that the test result of character, Fig. 7 and Fig. 8 respectively show GA algorithms and CGA algorithm optimizations
Numeral 0~9 test result figure, there is mistake at 11 in 300 class testing samples in GA algorithms in Fig. 6, and after CGA optimizations
Grader only there are 3 class mistakes, its character identification rate apparently higher than the former.Because alphabetic sorter and digital alphabet mix
The grouped data that grader needs is 24 classes and 34 classes respectively, and its test result figure is especially complex to be difficult to analyze, and the present embodiment is straight
Connect the form for table 2 by its result statistics, as can be seen from the table, recognition result of the CGA optimized algorithms for three class graders
Obviously higher than GA algorithms, and some advantages on average operating time, so the effect of optimization of the present embodiment is good, robustness
By force.
Referring to Fig. 9, embodiment 2, embodiment 2 and the Chinese character part for differing only in Beijing's car plate of embodiment 1
SIFT feature template matches figure, as shown in Figure 5.
From embodiment 1 and embodiment 2, according to Hanzi structure complexity, local feature abundant information, propose to adopt Chinese character
Feature extraction is carried out with SIFT local features operator, and is identified by the way of template matches, is gathered under different background
Car plate picture, tested on matlab softwares, show that Chinese Character Recognition rate can reach 100%;According to the letter of digital alphabet structure
It is single, alphanumeric characters are extracted using 13 point feature extraction methods, the identification of line character are entered using SVMs, for branch
Vector machine Parametric optimization problem is held, is known using the kernel parameter and penalty factor of Chaos Genetic Algorithm Support Vector Machines Optimized to improve
Not rate, and compared with traditional genetic algorithm, training adaptation degree change curve has been described above chaos proposed by the present invention and lost
The superiority of propagation algorithm, and identification of the SVM models trained to test data further specify that to Chaos Genetic Algorithm
The ability of Support Vector Machines Optimized parameter, it was demonstrated that the present invention Chinese character and digital alphabet of car plate are separated, respectively identify Chinese character and
Digital alphabet, recombinant form the number-plate number and know method for distinguishing, achieve significant effect.
Claims (8)
1. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm, including Chinese Character Recognition, the Chinese character
Identification includes two steps:Extract the key point that SIFT operators feature forms character feature;By the key of the character feature of formation
The template matches of point and car plate Chinese character carry out Chinese Character Recognition, it is characterized in that:Also include digital alphabet to identify, comprise the following steps that:
1) SIFT operator features are extracted, form the key point of character feature;
2) key point of the character feature of formation and the template matches of car plate Chinese character are subjected to Chinese Character Recognition;
3) alphanumeric characters are extracted;
4) identification of alphanumeric characters;
5) Chinese Character Recognition and digital Letter identification combine to form number-plate number identification.
2. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm as claimed in claim 1, it is special
Sign is:Step 3) the extraction alphanumeric characters use ten line-of-sight courses.
3. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm as claimed in claim 1, it is special
Sign is:The alphanumeric characters of the step 4) are identified using SVMs, are comprised the following steps:
(1) formula is seen below using Radial basis kernel function generation SVMs, Radial basis kernel function:
K(x,xi)=exp-| | x-xi||2/2σ2} (9)
In formula:Exp represents the exponential function using natural constant e the bottom of as;X represents that pixel is thinking direction coordinate;xiRepresent i-th
Coordinate of the pixel in x directions;σ represents nuclear parameter.
(2) optimized with Chaos Genetic Algorithm and determine penalty factor and nuclear parameter σ;
(3) vector machine classification design is supported, to determine three class graders;
(4) the three class graders that the penalty factor of step (2) optimization and nuclear parameter σ are imported to SVMs are identified.
4. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm as claimed in claim 3, it is special
Sign is:The step (2) optimizes penalty factor with Chaos Genetic Algorithm and nuclear parameter σ comprises the following steps:
1. determine fitness function
Using the Radial basis kernel function of formula (9) as fitness function;
2. Initial parameter sets
Determine operational factor and population scale M;
Using binary coding, it is integer to select C ∈ [1,100], σ ∈ [0.0001,1], penalty factor and nuclear parameter σ ginseng
Array represents that the population size of its initial population is 80 with 21 bits;
Maximum iteration is n=100;
Crossover probability PcFor 0.8;
Mutation probability
Wherein L is chromosome length;
3. Logistic chaotic maps produce initial population
Logistic chaotic maps, its generation initial population mapping equation are:
xn+1=μ xn(1-xn) n=0,1,2 ... (11)
In formula, xnThe relative number of n rear individual of iteration is represented, μ is controling parameter, is takenWhen, Equation Iterative motion rail
Mark will be in Complete Chaos motion state;
4. calculate individual adaptation degree using algorithm of support vector machine
Individual adaptation degree is defined as formula:
5. crossover operation is carried out to population
Generate individual random combine, the combination to each pairing two-by-two in population between one (0,1) at random by system first
Number, decide whether to intersect by crossover probability, if intersect, using mapping generation sequence through simply map after utilize Gaussian function
Count to determine crossover location, otherwise, lower a pair of the combinations of matching, all crossover locations can be determined by a chaos sequence;
6. mutation operator is carried out to current population
Decide whether to morph with mutation probability by the number that generates at random first, mapped to Logistic assign n it is small
The initial value of difference, n chaos sequence will be generated;
7. meet that stop condition then obtains the SVMs penalty factor of optimization and nuclear parameter σ combination parameter;
8. decode the penalty factor and nuclear parameter σ optimized;
9. if stop condition is unsatisfactory for for 4. individual adaptation degree that step calculates, it is necessary to reselect and calculate.
5. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm as claimed in claim 4, it is special
Sign is:4. individual adaptation degree that the step calculates is unsatisfactory for stop condition, needs to reselect and calculate, its step bag
Include:
A chaos intialization populations
An initial vector, X are randomly generated firsti=(xi1,xi2,…xim), by XiAccording to Logistic mapping equations obtain with
Machine initializes population:
<mrow>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>+</mo>
<mn>1</mn>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>=</mo>
<mfenced open='{' close=''>
<mtable>
<mtr>
<mtd>
<msub>
<mrow>
<mn>4</mn>
<mi>&mu;x</mi>
</mrow>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<mn>0.5</mn>
<mo>-</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mtd>
<mtd>
<mn>0</mn>
<mo>&le;</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>&le;</mo>
<mn>0.5</mn>
</mtd>
</mtr>
<mtr>
<mtd>
<mn>1</mn>
<mo>-</mo>
<msub>
<mrow>
<mn>4</mn>
<mi>&mu;x</mi>
</mrow>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mrow>
<mo>(</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>-</mo>
<mn>0.5</mn>
<mo>)</mo>
</mrow>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>-</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>)</mo>
</mrow>
</mtd>
<mtd>
<mn>0.5</mn>
<mo>&le;</mo>
<msub>
<mi>x</mi>
<mrow>
<mi>i</mi>
<mo>,</mo>
<mi>j</mi>
</mrow>
</msub>
<mo>&le;</mo>
<mn>1</mn>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>13</mn>
<mo>)</mo>
</mrow>
</mrow>
By N number of vectorial X1,X2,…XNThe span section of objective optimization function requirements initializaing variable is transformed to, is obtained N number of first
Beginning variable is designated as population at individual quantity as initial population;
B chaotic crossovers
Control the frequency of crossover operation with chaos sequence, concrete operations are as follows:Using chaotic crossover mapping equation:
xn+1=4xn(1-xn) (14)
Appoint and take an initial value x0, can produce a sequence of iterations, the characteristics of this sequence be changed between [0,1], it is clear that
Can be by xn+1As a random switching, work as xn+1During more than the value selected in advance, with regard to carrying out crossover operation;Conversely, then not
Intersect, i.e.,:
P is previously selected value in above formula, takes 0.5;
C chaotic mutations
In simple hereditary calculation method, mutation probability pmSelection be random, or by caused chaos sequence come control become
The progress of ETTHER-OR operation;
D replicates excellent individual in parent colony and new colony and enters the next generation.
6. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm as claimed in claim 3, it is special
Sign is:Step (3) the support vector cassification design is using one-to-one integrated mode, it is necessary to three class grader quantity of construction
For:
In formula:N is natural number, value 10.
7. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm as claimed in claim 1, it is special
Sign is:The step of step 1) extraction SIFT operator features, is as follows:
I structure metric space:Assuming that I (x, y) is input picture, G (x, y, σ) is the Gaussian function of mutative scale, then the image
Metric space function L (x, y, σ) definition is following formula:
L (x, y, σ)=G (x, y, σ) * I (x, y) (1)
In formula, σ is scale factor, and * is convolution algorithm, and (x, y) represents location of pixels coordinate in image;
Gaussian function G (x, y, σ) is defined as following formula:
<mrow>
<mi>G</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>,</mo>
<mi>&sigma;</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<msup>
<mrow>
<mn>2</mn>
<mi>&pi;&sigma;</mi>
</mrow>
<mn>2</mn>
</msup>
</mfrac>
<msup>
<mi>e</mi>
<mrow>
<mo>-</mo>
<mrow>
<mo>(</mo>
<msup>
<mi>x</mi>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mi>y</mi>
<mn>2</mn>
</msup>
<mo>)</mo>
</mrow>
<mo>/</mo>
<msup>
<mrow>
<mn>2</mn>
<mi>&sigma;</mi>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msup>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>2</mn>
<mo>)</mo>
</mrow>
</mrow>
In order to effectively detect stable extremal point in metric space, using difference of Gaussian function D (x, y, σ), it is by right
What the image subtraction that two adjacent yardsticks differ k times obtained:
D (x, y, σ)=(G (x, y, k σ)-G (x, y, σ)) * I (x, y)=L (x, y, k σ)-L (x, y, σ) (3)
II Local Extremum detects:In order to detect the local maximum of difference of Gaussian function and minimum, each sampled point will
Compared with it is with 8 consecutive points of yardstick and 9 × 2, totally 26 points of neighbouring yardstick, when the sampled point is big
When other, the sampled point is selected, therefore can detect Local Extremum all in the space,
As SIFT feature candidate point;
III is accurately positioned extreme point:The candidate's SIFT feature detected to previous step is screened, and filters out and noise is compared
The point sensitive, contrast is low and unstable marginal point, further to obtain the accurate point in position;
The screening technique of candidate's SIFT feature is to be fitted DOG spaces local sampling using three-dimensional quadratic equation, detection is simultaneously
Low contrast point is removed, carrying out Taylor expansion to quadratic term to D (x, y, σ) obtains:
<mrow>
<mi>D</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>D</mi>
<mo>+</mo>
<mfrac>
<mrow>
<mo>&part;</mo>
<msup>
<mi>D</mi>
<mi>T</mi>
</msup>
</mrow>
<mrow>
<mo>&part;</mo>
<mi>x</mi>
</mrow>
</mfrac>
<mo>+</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<msup>
<mi>x</mi>
<mi>T</mi>
</msup>
<mfrac>
<mrow>
<msup>
<mo>&part;</mo>
<mn>2</mn>
</msup>
<mi>D</mi>
</mrow>
<mrow>
<mo>&part;</mo>
<msup>
<mi>x</mi>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mi>x</mi>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>4</mn>
<mo>)</mo>
</mrow>
</mrow>
To above formula derivation, and make equation be equal to zero, D (x) extreme point can be obtainedIt is shown below:
<mrow>
<mover>
<mi>x</mi>
<mo>^</mo>
</mover>
<mo>=</mo>
<mo>-</mo>
<mfrac>
<mrow>
<msup>
<mo>&part;</mo>
<mn>2</mn>
</msup>
<msup>
<mi>D</mi>
<mrow>
<mo>-</mo>
<mn>1</mn>
</mrow>
</msup>
</mrow>
<mrow>
<mo>&part;</mo>
<msup>
<mi>x</mi>
<mn>2</mn>
</msup>
</mrow>
</mfrac>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>D</mi>
</mrow>
<mrow>
<mo>&part;</mo>
<mi>x</mi>
</mrow>
</mfrac>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>5</mn>
<mo>)</mo>
</mrow>
</mrow>
Formula (5) is substituted into formula (4) and obtains formula (6):
<mrow>
<mi>D</mi>
<mrow>
<mo>(</mo>
<mover>
<mi>x</mi>
<mo>^</mo>
</mover>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>D</mi>
<mo>+</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<mfrac>
<mrow>
<mo>&part;</mo>
<msup>
<mi>D</mi>
<mi>T</mi>
</msup>
</mrow>
<mrow>
<mo>&part;</mo>
<mi>x</mi>
</mrow>
</mfrac>
<mover>
<mi>x</mi>
<mo>^</mo>
</mover>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>6</mn>
<mo>)</mo>
</mrow>
</mrow>
If it is calculatedThen this feature point remains, and otherwise removes, and stays pinpoint extreme point just
It is key point;
IV key point direction is distributed and generation SIFT feature vector
, it is necessary to determine a main side of key point according to the size of its neighborhood territory pixel and directional spreding situation after key point determination
To realize the rotational invariance of SIFT operators;
Any one key point L (x, y) can calculate its gradient modulus value m (x, y) and direction θ (x, y), see formula (7)
(8):
<mrow>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msqrt>
<mrow>
<msup>
<mrow>
<mo>(</mo>
<mi>L</mi>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>+</mo>
<mn>1</mn>
<mo>,</mo>
<mi>y</mi>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mi>L</mi>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>-</mo>
<mn>1</mn>
<mo>,</mo>
<mi>y</mi>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<msup>
<mrow>
<mo>(</mo>
<mi>L</mi>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mi>L</mi>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
<mo>)</mo>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
</mrow>
</msqrt>
<mo>-</mo>
<mo>-</mo>
<mo>-</mo>
<mrow>
<mo>(</mo>
<mn>7</mn>
<mo>)</mo>
</mrow>
</mrow>
θ (x, y)=tan-1(L(x,y+1)-L(x.y-1))/(L(x+1,y)-L(x-1,y)) (8)
The gradient modulus value m (x, y) and direction θ (x, y) that are calculated according to formula (7) and (8) are the principal direction of key point.
8. a kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm as claimed in claim 1, it is special
Sign is:, will be each when the key point for the character feature that the step 2) is formed and the template matches of car plate Chinese character carry out Chinese Character Recognition
Data after the feature extraction of character are imported matlab and matched using nearest neighbor algorithm, when both matching numbers are most
When, corresponding car plate Chinese character is character to be identified.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710429590.3A CN107358231B (en) | 2017-06-09 | 2017-06-09 | License plate character recognition method based on SIFT operator and chaotic genetic algorithm |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710429590.3A CN107358231B (en) | 2017-06-09 | 2017-06-09 | License plate character recognition method based on SIFT operator and chaotic genetic algorithm |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107358231A true CN107358231A (en) | 2017-11-17 |
CN107358231B CN107358231B (en) | 2021-10-15 |
Family
ID=60272688
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710429590.3A Active CN107358231B (en) | 2017-06-09 | 2017-06-09 | License plate character recognition method based on SIFT operator and chaotic genetic algorithm |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107358231B (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110990353A (en) * | 2019-12-11 | 2020-04-10 | 深圳证券交易所 | Log extraction method, log extraction device and storage medium |
CN112085059A (en) * | 2020-08-06 | 2020-12-15 | 温州大学 | Breast cancer image feature selection method based on improved sine and cosine optimization algorithm |
CN112636874A (en) * | 2020-12-21 | 2021-04-09 | 西安理工大学 | Chaotic baseband wireless communication decoding method based on genetic optimization support vector machine |
CN113128768A (en) * | 2021-04-22 | 2021-07-16 | 中国电建集团中南勘测设计研究院有限公司 | Water-wind-fire short-term optimization scheduling method considering wind-electricity uncertainty |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101339601A (en) * | 2008-08-15 | 2009-01-07 | 张擎宇 | License plate Chinese character recognition method based on SIFT algorithm |
US20100169576A1 (en) * | 2008-12-31 | 2010-07-01 | Yurong Chen | System and method for sift implementation and optimization |
CN104834932A (en) * | 2015-04-29 | 2015-08-12 | 河南城建学院 | Matlab algorithm of automobile license plate identification |
CN105631445A (en) * | 2014-11-06 | 2016-06-01 | 通号通信信息集团有限公司 | Character recognition method and system for license plate with Chinese characters |
-
2017
- 2017-06-09 CN CN201710429590.3A patent/CN107358231B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101339601A (en) * | 2008-08-15 | 2009-01-07 | 张擎宇 | License plate Chinese character recognition method based on SIFT algorithm |
US20100169576A1 (en) * | 2008-12-31 | 2010-07-01 | Yurong Chen | System and method for sift implementation and optimization |
CN105631445A (en) * | 2014-11-06 | 2016-06-01 | 通号通信信息集团有限公司 | Character recognition method and system for license plate with Chinese characters |
CN104834932A (en) * | 2015-04-29 | 2015-08-12 | 河南城建学院 | Matlab algorithm of automobile license plate identification |
Non-Patent Citations (3)
Title |
---|
史海玲: "《运动模糊车牌识别关键技术研究》", 《中国优秀硕士学位论文全文数据库》 * |
周彬彬: "复杂环境下车牌识别系统中关键技术的研究与实现", 《中国优秀硕士学位论文全文数据库》 * |
郭金芝: "基于SIFT算法的车牌识别系统研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110990353A (en) * | 2019-12-11 | 2020-04-10 | 深圳证券交易所 | Log extraction method, log extraction device and storage medium |
CN110990353B (en) * | 2019-12-11 | 2023-10-13 | 深圳证券交易所 | Log extraction method, log extraction device and storage medium |
CN112085059A (en) * | 2020-08-06 | 2020-12-15 | 温州大学 | Breast cancer image feature selection method based on improved sine and cosine optimization algorithm |
CN112085059B (en) * | 2020-08-06 | 2023-10-20 | 温州大学 | Breast cancer image feature selection method based on improved sine and cosine optimization algorithm |
CN112636874A (en) * | 2020-12-21 | 2021-04-09 | 西安理工大学 | Chaotic baseband wireless communication decoding method based on genetic optimization support vector machine |
CN112636874B (en) * | 2020-12-21 | 2022-08-26 | 西安理工大学 | Chaotic baseband wireless communication decoding method based on genetic optimization support vector machine |
CN113128768A (en) * | 2021-04-22 | 2021-07-16 | 中国电建集团中南勘测设计研究院有限公司 | Water-wind-fire short-term optimization scheduling method considering wind-electricity uncertainty |
Also Published As
Publication number | Publication date |
---|---|
CN107358231B (en) | 2021-10-15 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107358231A (en) | A kind of license plate character recognition method based on SIFT operators and Chaos Genetic Algorithm | |
Li et al. | DeepMitosis: Mitosis detection via deep detection, verification and segmentation networks | |
CN102054178B (en) | A kind of image of Chinese Painting recognition methods based on local semantic concept | |
CN108537775A (en) | A kind of cancer cell tracking based on deep learning detection | |
CN109000645A (en) | Complex environment target classics track extracting method | |
CN112016605B (en) | Target detection method based on corner alignment and boundary matching of bounding box | |
CN105046673B (en) | High spectrum image and visual image fusion sorting technique based on self study | |
CN103390169A (en) | Sorting method of vehicle-mounted laser scanning point cloud data of urban ground objects | |
CN107247956A (en) | A kind of fast target detection method judged based on grid | |
CN105005764A (en) | Multi-direction text detection method of natural scene | |
CN105975968A (en) | Caffe architecture based deep learning license plate character recognition method | |
CN103324939B (en) | Skewed popularity classification and parameter optimization method based on least square method supporting vector machine technology | |
CN105513066B (en) | It is a kind of that the generic object detection method merged with super-pixel is chosen based on seed point | |
CN111832615A (en) | Sample expansion method and system based on foreground and background feature fusion | |
CN103413145A (en) | Articulation point positioning method based on depth image | |
CN107273832A (en) | Licence plate recognition method and system based on integrating channel feature and convolutional neural networks | |
CN107169496A (en) | A kind of character recognition method and device | |
US11663499B2 (en) | Method, data processing apparatus and computer program product for determining road intersections | |
CN112528934A (en) | Improved YOLOv3 traffic sign detection method based on multi-scale feature layer | |
CN113223068A (en) | Multi-modal image registration method and system based on depth global features | |
CN107861957A (en) | A kind of data analysing method and device | |
CN112560915A (en) | Urban expressway traffic state identification method based on machine learning | |
CN106933977A (en) | It is a kind of that the method that flight parameter outlier is rejected in classification is excavated based on big data | |
CN111242144A (en) | Method and device for detecting abnormality of power grid equipment | |
CN104573683A (en) | Character string recognizing method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |