CN113554028A - License plate character classification method - Google Patents
License plate character classification method Download PDFInfo
- Publication number
- CN113554028A CN113554028A CN202110966757.6A CN202110966757A CN113554028A CN 113554028 A CN113554028 A CN 113554028A CN 202110966757 A CN202110966757 A CN 202110966757A CN 113554028 A CN113554028 A CN 113554028A
- Authority
- CN
- China
- Prior art keywords
- dimension
- license plate
- matrix
- reduction
- characters
- 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.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Character Discrimination (AREA)
Abstract
A license plate character classification method relates to the field of character classification, and comprises the following steps: s1, obtaining a training set dimension reduction matrix under each dimension reduction dimension through an SVD algorithm; and obtaining the dimension reduction matrix of the test set under each dimension reduction dimension based on the process of obtaining the dimension reduction matrix of each training set. And S2, carrying out classification training on the vectors of the dimensionality reduction matrix of each training set through an SVM algorithm to obtain a multi-class classifier of license plate characters under each dimensionality reduction dimension. S3, determining an optimal multi-class classifier. And S4, converting the license plate character image to be recognized into a dimension reduction vector to be recognized, and predicting the character class to which the vector belongs by using the optimal multi-class classifier. The SVD algorithm reduces the dimension of the original matrix of the training set, and simplifies the parameters required for obtaining various multi-class classifiers. The obtained optimal multi-class classifier has low time cost and space cost and high classification precision, and can be widely applied to embedded equipment with limited resources.
Description
Technical Field
The invention relates to the field of character classification, in particular to a license plate character classification method.
Background
In the urban development process, the automobile holding capacity is continuously improved, and with the popularization of artificial intelligence technology, automatic management technology permeates into automobile management and other related industries. To realize an automatic management mode of automobiles, the identity "ID", i.e., the license plate, of each automobile needs to be recognized first.
The currently popular character recognition methods include the following: 1. a classification method based on template matching; 2. a neural network based character classification method; 3. character classification method based on SVM.
The template matching-based classification method is low in cost and high in recognition speed, and basically meets the requirement of real-time performance, but the template matching mode has extremely high requirements on the license plate extraction and character segmentation effects, and the accuracy of matching approximate characters such as 0 and D, 0 and Q is not high. The character classification algorithm based on the neural network relies on the network structure design, and the neural network parameter storage capacity is larger, so that the method is not suitable for embedded equipment for recognizing license plates. The character classification method based on the SVM is simple to implement, and can achieve high classification accuracy under the condition of not considering training cost, but the existing scheme mostly uses original character image data, and has high calculation dimensionality, so that an SVM model is complex, the data storage capacity is large, and the time cost of data classification is increased.
For example, the chinese invention with application number CN200910059360.8 discloses a license plate character recognition method based on multi-classification support vector machines, which divides the feature vectors of the binary image of the license plate characters into four sets, respectively constructs four multi-classification support vector machines to calculate the support vector sets of the four sets, and adopts different support vector sets to recognize the characters of the license plate to be recognized one by one in the license plate recognition process. The method only binarizes the original character image, has high calculation dimensionality, and is difficult to finish character recognition with high precision under the conditions of small data storage capacity and low time cost.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a license plate character recognition method, which can ensure the classification precision of license plate characters under low-dimensional characteristics and can be widely applied to embedded equipment with limited resources in the aspects of storage space and the like.
The invention is realized by the following technical scheme:
a method for classifying characters of a license plate comprises the following steps:
step S1, performing dimension reduction of different dimension reduction dimensions on the original matrix of the training set of the license plate characters through a Singular Value Decomposition (SVD) algorithm to obtain a training set dimension reduction matrix of the license plate characters under each dimension reduction dimension; and performing dimension reduction of different dimension reduction on the original matrix of the test set of the license plate characters based on the dimension reduction process of the original matrix of the training set of the license plate characters to obtain the dimension reduction matrix of the test set of the license plate characters under each dimension reduction.
And step S2, respectively carrying out classification training on vectors in the dimension reduction matrix of the training set of the license plate characters under each dimension reduction dimension through a Support Vector Machine (SVM) algorithm to obtain a multi-class classifier of the license plate characters under each dimension reduction dimension.
And step S3, classifying each vector in the dimension reduction matrix of the test set under the same dimension reduction dimension by using a multi-class classifier of the license plate characters under each dimension reduction dimension, and determining one multi-class classifier under the dimension reduction dimension as the optimal multi-class classifier according to the space cost, the time cost and the classification precision of each multi-class classifier.
And step S4, converting the license plate character image to be recognized into a dimension reduction vector to be recognized, wherein the dimension reduction vector is the same as the dimension reduction dimension of the optimal multi-class classifier, and predicting the character class to which the dimension reduction vector to be recognized belongs by utilizing the optimal multi-class classifier.
The SVD algorithm reduces the dimension of the original matrix of the training set, simplifies the parameters required for obtaining the multi-class classifiers under each dimension reduction dimension, and further enables the space cost of each multi-class classifier to be lower. The multi-class classifiers classify the dimension reduction matrixes of the test set under the same dimension reduction dimension respectively, and then the multi-class classifier under one dimension reduction dimension is determined to be the best multi-class classifier according to the space cost, the time cost and the classification precision of the multi-class classifiers, so that the multi-class classifier can be more widely applied to embedded equipment with limited resources in the aspects of storage space and the like.
Preferably, before step S1, the method further includes: and step S01, collecting a plurality of original images of various characters for forming the license plate number to form an original image set. Step S02, preprocessing each image in the original image set to obtain a preprocessed image set, dividing the images in the preprocessed image set into a training set and a test set, wherein the ratio of the number of each type of character image in the test set to the number of the corresponding type of character images in the training set is the same. Step S03, converting each image in the training set into a training image matrix by using an image conversion matrix function, converting each training image matrix into a training vector by using a vector conversion function, and forming all training vectors converted from all images in the training set into a training set original matrix of license plate characters; and converting each image in the test set into a test image matrix by using an image conversion matrix function, converting each test image matrix into a test vector by using a vector conversion function, and converting all test vectors of all images in the test set into a test set original matrix of the license plate characters.
Preferably, the step S1 specifically includes: s11, performing singular value decomposition on a training set original matrix of license plate characters by using a Singular Value Decomposition (SVD) algorithm, selecting front n-dimensional feature vectors in a left singular matrix obtained after singular value decomposition, wherein n is a natural number which is more than or equal to 1 and less than or equal to x, and respectively forming each low-dimensional space coordinate system under the condition that n is different values; wherein x is the resolution of any image in the training set in the step S02, and the size and resolution of each image in the training set in the step S02 are the same; step S12, mapping the original matrix of the training set of the license plate characters to each low-dimensional space coordinate system in the step S11 respectively to obtain a training set dimension reduction matrix of the license plate characters under each dimension reduction dimension; and respectively mapping the original matrix of the test set of the license plate characters to each low-dimensional space coordinate system in the step S11 to obtain a dimension reduction matrix of the test set of the license plate characters under each dimension reduction dimension.
Preferably, in step S2, a one-to-one multi-class classification method is selected to construct a multi-class classifier under each dimensionality reduction dimension.
Preferably, the one-to-one multi-class classification method specifically includes: in a training set dimension reduction matrix of license plate characters under any dimension reduction dimension, taking any two types of character vectors to design a character sub-classifier; when the character category number is k, k x (k-1)/2 character sub-classifiers are designed under each dimensionality reduction to form a multi-category classifier of license plate characters under each dimensionality reduction, wherein k is a natural number larger than 1.
Preferably, the multi-class classifiers in each dimension reduction dimension are composed of a plurality of character sub-classifiers, and the step S3 specifically includes: using each character sub-classifier in the multi-class classifiers under each dimensionality reduction dimension to classify the vectors in the dimensionality reduction matrix of the test set of the license plate characters under the same dimensionality reduction dimension as the multi-class classifiers under each dimensionality reduction dimension, and taking the classification result with the highest occurrence frequency in the classification results obtained by each character sub-classifier in each multi-class classifier as a final prediction result; and then determining the multi-class classifier under one dimension reduction dimension as the optimal multi-class classifier according to the space cost, the time cost and the classification precision of the multi-class classifier under each dimension reduction dimension.
Preferably, the multi-class classifier under each dimensionality reduction dimension respectively comprises a plurality of character sub-classifiers, and the determination process of the space cost, the time cost and the classification precision of each multi-class classifier specifically comprises the following steps: the calculation mode of the space cost p of each multi-class classifier is as follows: p = n × m; the time cost t of each multi-class classifier is calculated in the following mode: t = n m + n x; where n is a dimension reduction dimension, m is the number of character sub-classifiers included in each multi-class classifier, and x is the resolution of any image in the training set in step S02. The obtaining mode of the classification precision of each multi-class classifier is as follows: and calculating the ratio of the number of vectors correctly classified by each multi-class classifier to the number of vectors for classification by using a classification precision function so as to obtain the classification precision of each multi-class classifier.
Preferably, the step S4 specifically includes: step S41, preprocessing the license plate character image to be recognized to obtain a preprocessed image to be recognized, converting the preprocessed image to be recognized into a preprocessed image matrix to be recognized by using an image conversion matrix function, and converting the preprocessed image matrix to be recognized into a vector to be recognized by using a vector conversion function; and carrying out dimensionality reduction on the vector to be identified, wherein the dimensionality reduction is the same as that of a dimensionality reduction matrix of a training set of the optimal multi-class classifier obtained by training, and obtaining the dimensionality reduction vector to be identified. And step S42, classifying the dimensionality reduction vector to be recognized by using the optimal multi-class classifier to obtain a prediction result of the character class of the dimensionality reduction vector to be recognized.
Preferably, the image preprocessing process includes: processing the size and resolution of the image to unify the size and resolution of all the images; and then, sequentially performing color removal, noise reduction and binarization processing on each image.
Preferably, in the step S01, the types of characters used to form the license plate number are a total of 34 types of license plate characters including numeric characters 0 to 9 and alphabetic characters a-H, J-N, P-Z.
The invention has the following beneficial effects:
1. the invention reduces dimensions of different dimension reduction dimensions on the original matrix of the training set of license plate characters through the singular value decomposition SVD algorithm, simplifies parameters required by obtaining multi-class classifiers under each dimension reduction dimension, and further reduces the space cost of each multi-class classifier. The vectors in the dimensionality reduction matrix of the test set under the same dimensionality reduction dimension are classified through each multi-class classifier, the time cost and the classification precision of each multi-class classifier are counted, and the optimal multi-class classifier is determined by combining the space cost of each multi-class classifier, so that the method can be widely applied to embedded equipment with limited resources.
2. The multi-class classifier is obtained through the SVM algorithm, and compared with the classifier obtained through the neural network algorithm with the classification precision depending on the network structure design, the classifier obtained through the SVM algorithm is more suitable for embedded equipment with relatively limited resources.
3. Compared with a one-to-many classification method, the invention adopts the one-to-one multi-class classification method to construct the multi-class classifier under each dimensionality reduction dimension, can avoid the problem of data imbalance, has better classification boundary, and does not need to retrain the whole classification model after introducing a new class.
Drawings
FIG. 1 is a process flow diagram.
Detailed Description
The following is a specific embodiment of the present invention, and the technical solution of the present invention is further described with reference to the accompanying drawings, but the present invention is not limited to the embodiment, and after reading the present specification, a person skilled in the art may make modifications to the embodiment as needed without inventive contribution, but the invention is protected by patent laws within the scope of the claims of the present invention.
In an embodiment, as shown in fig. 1, a method for classifying characters of a license plate includes:
step S1, performing dimension reduction of different dimension reduction dimensions on the original matrix of the training set of the license plate characters through a Singular Value Decomposition (SVD) algorithm to obtain a training set dimension reduction matrix of the license plate characters under each dimension reduction dimension; and performing dimension reduction of different dimension reduction on the original matrix of the test set of the license plate characters based on the dimension reduction process of the original matrix of the training set of the license plate characters to obtain the dimension reduction matrix of the test set of the license plate characters under each dimension reduction.
And step S2, respectively carrying out classification training on vectors in the dimension reduction matrix of the training set of the license plate characters under each dimension reduction dimension through a Support Vector Machine (SVM) algorithm to obtain a multi-class classifier of the license plate characters under each dimension reduction dimension.
And step S3, classifying each vector in the dimension reduction matrix of the test set under the same dimension reduction dimension by using a multi-class classifier of the license plate characters under each dimension reduction dimension, and determining one multi-class classifier under the dimension reduction dimension as the optimal multi-class classifier according to the space cost, the time cost and the classification precision of each multi-class classifier.
And step S4, converting the license plate character image to be recognized into a dimension reduction vector to be recognized, wherein the dimension reduction vector is the same as the dimension reduction dimension of the optimal multi-class classifier, and predicting the character class to which the dimension reduction vector to be recognized belongs by utilizing the optimal multi-class classifier.
Wherein, the step S1 is preceded by a step S01, a step S02 and a step S03. Step S01: a plurality of original images of various types of characters for composing a license plate number are collected, such as 13163 random original images of 34 types of license plate characters in total, including numerical characters 0 to 9 and alphabetic characters A-H, J-N, P-Z, to compose an original image set of license plate characters. Step S02: importing each image in an original image set of license plate characters into matlab software for preprocessing to obtain a preprocessed image set, wherein the preprocessing process of each image comprises the following steps: unifying the sizes and resolutions of all images, and then sequentially performing decoloring, denoising and binarization processing on each image. Then, images in the preprocessed image set are divided into a training set and a testing set of license plate characters according to the ratio of 8 to 2, and the ratio of the number of various types of character images in the training set to the number of corresponding types of character images in the testing set is also 8 to 2. Step S03: and converting each image in the training set into a training image matrix by using an image conversion matrix function, such as an imread function of matlab software, converting each training image matrix into a training vector by using a vector conversion function, such as a reshape function of matlab software, and forming all training vectors converted from all images in the training set into a training set original matrix. And then, converting each image in the test set into a test image matrix by using an imread function of matlab software, converting each test image matrix into a test vector by using a reshape function of the matlab software, and forming an original matrix of the test set by using all the test vectors after all the images in the test set are converted.
The step S1 is to perform dimension reduction through a singular value decomposition SVD algorithm, and the step S1 specifically includes steps S11 and S12. Step S11: and carrying out singular value decomposition on the original matrix A of the training set by using a singular value decomposition SVD algorithm to obtain decomposition results U, sigma and V, wherein U and V are unit orthogonal arrays, U is a left singular matrix, V is a right singular matrix, and sigma is a singular value matrix. The column vectors in U and V are respectively used as base vectors of a column space and a row space of A, in a training set original matrix A, the column space represents characteristics, and the row space represents a sample. Then, the front n-dimensional feature vectors of the left singular matrix U are selected to respectively form each low-dimensional space coordinate system under the condition that n is different values, and n is a natural number which is greater than or equal to 1 and less than or equal to x, wherein x is the resolution of any image in the training set in the step S02, and if the resolutions of any images in the training set are unified to be 20 × 20, x is 400. Step S12: mapping the each vector in the original matrix of the training set to each low-dimensional space coordinate system in the step S11 respectively to obtain a training set dimension reduction matrix under each dimension reduction; and mapping the original matrix of the test set to each low-dimensional space coordinate system in the step S11 respectively to obtain the dimension reduction matrix of the test set under each dimension reduction dimension.
And step S2, obtaining a multi-class classifier of the license plate characters under each dimension reduction dimension through a Support Vector Machine (SVM) algorithm. In step S2, a one-to-one multi-class classification method is selected to construct a multi-class classifier under each dimensionality reduction dimension. And each multi-class classifier consists of a plurality of character sub-classifiers. The one-to-one multi-class classification method specifically comprises the following steps: in a training set dimension reduction matrix under any dimension reduction dimension, taking any two types of character vectors to design a character sub-classifier; when the number of character categories contained in the training set dimension reduction matrix is k, k x (k-1)/2 character sub-classifiers in each dimension reduction dimension are designed to form a multi-category classifier in each dimension reduction dimension, wherein k is a natural number larger than 1. If the present invention includes 34 types of characters in total including numeric characters 0 to 9 and alphabetic characters A-H, J-N, P-Z, then in each training set dimension reduction matrix, the vectors with character categories A and B, A, C, A and D … … are respectively selected to design character sub-classifiers, so as to obtain 561 character sub-classifiers in each dimension reduction. The multi-class classifier under each dimension reduction dimension respectively consists of character sub-classifiers under each dimension reduction dimension, the number of the license plate character classes of the invention is 34, and each multi-class classifier respectively comprises 561 character sub-classifiers.
The step S3 is used to determine the best multi-class classifier. The step S3 specifically includes: and (3) using each character sub-classifier in the multi-class classifiers under each dimensionality reduction dimension to classify each vector in the dimensionality reduction matrix of the test set under the same dimensionality reduction dimension as each multi-class classifier under each dimensionality reduction dimension, and taking the classification result with the highest occurrence frequency in the classification results obtained by each character sub-classifier in the multi-class classifiers as a final prediction classification result. And then determining the multi-class classifier under one dimension reduction dimension as the optimal multi-class classifier according to the space cost, the time cost and the classification precision of the multi-class classifier under each dimension reduction dimension. The determination process of the space cost, the time cost and the classification precision of each multi-class classifier specifically comprises the following steps: the calculation mode of the space cost p of each multi-class classifier is as follows: p = n × m; the time cost t of each multi-class classifier is calculated in the following mode: t = n m + n x; wherein n is a dimensionality reduction dimension, a value of the n corresponds to a value of n in step S11, m is the number of character sub-classifiers included in each multi-class classifier, m = k × (k-1)/2, and x is a resolution of any image in the training set. The obtaining mode of the classification precision of each multi-class classifier is as follows: the classification precision function, such as a matlab confusion matrix function, is utilized to obtain the ratio of the number of correctly classified vectors of each multi-class classifier to the number of classified vectors, so as to obtain the classification precision of each multi-class classifier, such as 13163 vectors which are classified in total, wherein the number of the vectors which are classified as the same as the number of the originally-belonging characters of the vector is 13093, and the classification precision is 13093/13163, that is, the classification precision is 99.47%.
In the invention, x is 400, and m is 561. When n takes on values of 8, 16, 24, 32, 96, 128 and 192, the statistical graph of the space cost, the time cost and the classification precision of the multi-class classifier is as follows:
statistical table of time cost, space cost and classification precision of multiclass classifier obtained by taking different values of table one n
As can be seen from the above table, when n is 32, the precision can reach ninety-nine percent or more, and when n continues to increase, the precision is not significantly improved, but the time cost and the space cost are greatly improved, so that the multi-class classifier when n is 32 is selected as the optimal multi-class classifier.
The step S4 is to predict the character classification to which the dimensionality reduction vector to be recognized belongs by using the optimal multi-class classifier. The step S4 specifically includes a step S41 and a step S42. Step S41: adjusting license plate character images to be recognized into images with the same size and resolution as those of the images in the preprocessed image set by using matlab software, then performing decoloration, noise reduction and binarization processing to obtain preprocessed images to be recognized, then converting the preprocessed images to be recognized into preprocessed image matrixes to be recognized by using an imread function of the matlab software, and then converting the preprocessed image matrixes to be recognized into vectors to be recognized by using a reshape function of the matlab software; and mapping the vector to be identified to a low-dimensional space coordinate system when n is 32 in the step S11 to obtain the reduced-dimension vector to be identified. Step S42: and classifying the dimensionality reduction vectors to be identified by using each character sub-classifier contained in the optimal multi-class classifier, and taking the classification result with the highest occurrence frequency in the classification results obtained by each character sub-classifier as the prediction result of the character classification of the dimensionality reduction vectors to be identified.
It will be appreciated by persons skilled in the art that the embodiments of the invention described above and shown in the drawings are given by way of example only and are not limiting of the invention. Those skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope of the present disclosure, and such modifications or substitutions are intended to be included within the scope of the present disclosure. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims. The objects of the present invention have been fully and effectively accomplished. The functional and structural principles of the present invention have been shown and described in the examples, and any variations or modifications of the embodiments of the present invention may be made without departing from the principles.
Claims (10)
1. A method for classifying characters of a license plate is characterized by comprising the following steps:
step S1, performing dimension reduction of different dimension reduction dimensions on the original matrix of the training set of the license plate characters through a Singular Value Decomposition (SVD) algorithm to obtain a training set dimension reduction matrix of the license plate characters under each dimension reduction dimension; based on the dimension reduction process of the original matrix of the training set of the license plate characters, carrying out dimension reduction of different dimension reduction on the original matrix of the testing set of the license plate characters to obtain the dimension reduction matrix of the testing set of the license plate characters under each dimension reduction;
step S2, respectively carrying out classification training on vectors in a dimension reduction matrix of a training set of license plate characters under each dimension reduction through a Support Vector Machine (SVM) algorithm to obtain a multi-class classifier of the license plate characters under each dimension reduction;
step S3, using multi-class classifiers of license plate characters under each dimensionality reduction dimension to classify each vector in a dimensionality reduction matrix of a test set under the same dimensionality reduction dimension, and then determining one multi-class classifier under one dimensionality reduction dimension as an optimal multi-class classifier according to the space cost, the time cost and the classification precision of each multi-class classifier;
and step S4, converting the license plate character image to be recognized into a dimension reduction vector to be recognized, wherein the dimension reduction vector is the same as the dimension reduction dimension of the optimal multi-class classifier, and predicting the character class to which the dimension reduction vector to be recognized belongs by utilizing the optimal multi-class classifier.
2. The method for classifying characters on a license plate of claim 1, wherein before the step S1, the method further comprises:
step S01, collecting a plurality of original images of various characters for forming the license plate number to form an original image set;
step S02, preprocessing each image in the original image set to obtain a preprocessed image set, dividing the images in the preprocessed image set into a training set and a test set, wherein the ratio of the number of each type of character images in the test set to the number of the corresponding type of character images in the training set is the same;
step S03, converting each image in the training set into a training image matrix by using an image conversion matrix function, converting each training image matrix into a training vector by using a vector conversion function, and forming all training vectors converted from all images in the training set into a training set original matrix of license plate characters; and converting each image in the test set into a test image matrix by using an image conversion matrix function, converting each test image matrix into a test vector by using a vector conversion function, and converting all test vectors of all images in the test set into a test set original matrix of the license plate characters.
3. The method for classifying characters on a license plate according to claim 2, wherein the step S1 specifically includes:
s11, performing singular value decomposition on a training set original matrix of license plate characters by using a Singular Value Decomposition (SVD) algorithm, selecting front n-dimensional feature vectors in a left singular matrix obtained after singular value decomposition, wherein n is a natural number which is more than or equal to 1 and less than or equal to x, and respectively forming each low-dimensional space coordinate system under the condition that n is different values; wherein x is the resolution of any image in the training set in the step S02, and the size and resolution of each image in the training set in the step S02 are the same;
step S12, mapping the original matrix of the training set of the license plate characters to each low-dimensional space coordinate system in the step S11 respectively to obtain a training set dimension reduction matrix of the license plate characters under each dimension reduction dimension; and respectively mapping the original matrix of the test set of the license plate characters to each low-dimensional space coordinate system in the step S11 to obtain a dimension reduction matrix of the test set of the license plate characters under each dimension reduction dimension.
4. The method for classifying characters on license plates according to claim 1, wherein said step S2 employs a one-to-one multi-class classification method to construct multi-class classifiers for each dimension reduction.
5. The method for classifying characters on a license plate according to claim 4, wherein the one-to-one multi-class classification method is specifically: in a training set dimension reduction matrix of license plate characters under any dimension reduction dimension, taking any two types of character vectors to design a character sub-classifier; when the character category number is k, k x (k-1)/2 character sub-classifiers are designed under each dimensionality reduction to form a multi-category classifier of license plate characters under each dimensionality reduction, wherein k is a natural number larger than 1.
6. The method for classifying characters on a license plate according to claim 1, wherein each of the multiple classifiers in each dimension-reduction dimension is composed of multiple character sub-classifiers, and the step S3 specifically includes: using each character sub-classifier in the multi-class classifiers under each dimensionality reduction dimension to classify the vectors in the dimensionality reduction matrix of the test set of the license plate characters under the same dimensionality reduction dimension as the multi-class classifiers under each dimensionality reduction dimension, and taking the classification result with the highest occurrence frequency in the classification results obtained by each character sub-classifier in each multi-class classifier as a final prediction result; and then determining the multi-class classifier under one dimension reduction dimension as the optimal multi-class classifier according to the space cost, the time cost and the classification precision of the multi-class classifier under each dimension reduction dimension.
7. The method for classifying characters on a license plate according to claim 2, wherein the multi-class classifiers under each dimensionality reduction respectively comprise a plurality of character sub-classifiers, and the determination process of the space cost, the time cost and the classification precision of each multi-class classifier specifically comprises:
the calculation mode of the space cost p of each multi-class classifier is as follows: p = n × m;
the time cost t of each multi-class classifier is calculated in the following mode: t = n m + n x;
wherein n is a dimension reduction dimension, m is the number of character sub-classifiers included in each multi-class classifier, and x is the resolution of any image in the training set in the step S02;
the obtaining mode of the classification precision of each multi-class classifier is as follows: and calculating the ratio of the number of vectors correctly classified by each multi-class classifier to the number of vectors for classification by using a classification precision function so as to obtain the classification precision of each multi-class classifier.
8. The method for classifying characters on a license plate according to claim 1, wherein the step S4 specifically includes:
step S41, preprocessing the license plate character image to be recognized to obtain a preprocessed image to be recognized, converting the preprocessed image to be recognized into a preprocessed image matrix to be recognized by using an image conversion matrix function, and converting the preprocessed image matrix to be recognized into a vector to be recognized by using a vector conversion function; carrying out dimensionality reduction on the vector to be identified, wherein the dimensionality reduction is the same as that of a dimensionality reduction matrix of a training set of an optimal multi-class classifier obtained through training, and obtaining the dimensionality reduction vector to be identified;
and step S42, classifying the dimensionality reduction vector to be recognized by using the optimal multi-class classifier to obtain a prediction result of the character class of the dimensionality reduction vector to be recognized.
9. The method for classifying characters on a license plate according to claim 2 or 8, wherein the preprocessing of the image comprises: processing the size and resolution of the image to unify the size and resolution of all the images; and then, sequentially performing color removal, noise reduction and binarization processing on each image.
10. The method of claim 2, wherein in step S01, the types of characters used to form the license plate number are a total of 34 types of license plate characters including numeric characters 0 to 9 and alphabetic characters a-H, J-N, P-Z.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110966757.6A CN113554028A (en) | 2021-08-23 | 2021-08-23 | License plate character classification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110966757.6A CN113554028A (en) | 2021-08-23 | 2021-08-23 | License plate character classification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113554028A true CN113554028A (en) | 2021-10-26 |
Family
ID=78134141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110966757.6A Pending CN113554028A (en) | 2021-08-23 | 2021-08-23 | License plate character classification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113554028A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114387619A (en) * | 2021-12-31 | 2022-04-22 | 歌尔科技有限公司 | Pedestrian detection method, device, electronic equipment and computer-readable storage medium |
-
2021
- 2021-08-23 CN CN202110966757.6A patent/CN113554028A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114387619A (en) * | 2021-12-31 | 2022-04-22 | 歌尔科技有限公司 | Pedestrian detection method, device, electronic equipment and computer-readable storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chacko et al. | Handwritten character recognition using wavelet energy and extreme learning machine | |
| CN102982349A (en) | Image recognition method and device | |
| CN101739555B (en) | Method and system for detecting false face, and method and system for training false face model | |
| CN110717554B (en) | Image recognition method, electronic device, and storage medium | |
| CN1908960A (en) | Feature classification based multiple classifiers combined people face recognition method | |
| CN109165639B (en) | Finger vein identification method, device and equipment | |
| CN107292225B (en) | Face recognition method | |
| CN102156871A (en) | Image classification method based on category correlated codebook and classifier voting strategy | |
| Zhong et al. | A comparative study of image classification algorithms for Foraminifera identification | |
| CN109325507A (en) | A kind of image classification algorithms and system of combination super-pixel significant characteristics and HOG feature | |
| Zhang et al. | Automatic discrimination of text and non-text natural images | |
| Lai et al. | Learning discriminative feature hierarchies for off-line signature verification | |
| CN109858518A (en) | A kind of large data clustering method based on MapReduce | |
| Zhang et al. | Large-scale aerial image categorization using a multitask topological codebook | |
| CN105023006A (en) | Face recognition method based on enhanced nonparametric margin maximization criteria | |
| CN113489685A (en) | Secondary feature extraction and malicious attack identification method based on kernel principal component analysis | |
| Tsourounis et al. | Handwritten signature verification via deep sparse coding architecture | |
| CN113554028A (en) | License plate character classification method | |
| Liu et al. | Language identification of character images using machine learning techniques | |
| CN112070116A (en) | Automatic art painting classification system and method based on support vector machine | |
| CN104166837B (en) | Using the visual speech recognition methods of the selection of each group of maximally related point of interest | |
| CN107563287B (en) | Face recognition method and device | |
| CN111931229B (en) | Data identification method, device and storage medium | |
| CN112257792A (en) | SVM (support vector machine) -based real-time video target dynamic classification method | |
| John et al. | Handwritten Malayalam character recognition with a novel gradient based feature descriptor and a comparative study using SVM and ELM |
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 |