CN112926610A - Construction method of license plate image screening model and license plate image screening method - Google Patents
Construction method of license plate image screening model and license plate image screening method Download PDFInfo
- Publication number
- CN112926610A CN112926610A CN201911239521.1A CN201911239521A CN112926610A CN 112926610 A CN112926610 A CN 112926610A CN 201911239521 A CN201911239521 A CN 201911239521A CN 112926610 A CN112926610 A CN 112926610A
- Authority
- CN
- China
- Prior art keywords
- license plate
- plate image
- data
- model
- character recognition
- 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
- 238000012216 screening Methods 0.000 title claims abstract description 119
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000010276 construction Methods 0.000 title claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 47
- 238000009826 distribution Methods 0.000 claims abstract description 45
- 238000000605 extraction Methods 0.000 claims abstract description 25
- 238000003860 storage Methods 0.000 claims abstract description 16
- 238000012545 processing Methods 0.000 claims description 61
- 239000013598 vector Substances 0.000 claims description 28
- 230000006870 function Effects 0.000 claims description 25
- 238000010586 diagram Methods 0.000 claims description 20
- 238000004590 computer program Methods 0.000 claims description 19
- 238000013528 artificial neural network Methods 0.000 claims description 11
- 238000010606 normalization Methods 0.000 claims description 10
- 230000009467 reduction Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000008676 import Effects 0.000 claims description 3
- 230000002159 abnormal effect Effects 0.000 abstract description 3
- 238000004422 calculation algorithm Methods 0.000 description 30
- 230000000694 effects Effects 0.000 description 5
- 238000001514 detection method Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 230000004913 activation Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 238000011176 pooling Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012015 optical character recognition Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/217—Validation; Performance evaluation; Active pattern learning techniques
- G06F18/2193—Validation; Performance evaluation; Active pattern learning techniques based on specific statistical tests
-
- 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
- G06V20/625—License plates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Abstract
The application relates to a construction method and a device of a license plate image screening model, a storage medium, computer equipment, a license plate image screening method, a device of the license plate image screening model, a storage medium and computer equipment, wherein the construction method of the license plate image screening model comprises the following steps: obtaining error distribution sample data of the license plate image, performing feature extraction on the error distribution sample data of the license plate image, performing model training on a preset identification network through the extracted features, adjusting a cross entropy loss function of the identification network according to a model training result until the cross entropy loss function is minimized, and obtaining a license plate image screening model. The error distribution of the license plate image is learned through the error distribution sample data of the license plate image, subsequently, when the license plate is identified, the license plate image is firstly screened through a license plate image screening model, abnormal data in the license plate image is filtered through a front screening mode, and high-precision identification of the license plate at the later stage is supported.
Description
Technical Field
The application relates to the technical field of image processing, in particular to a construction method and device of a license plate image screening model, a storage medium, computer equipment, a license plate image screening method and device, a storage medium and computer equipment.
Background
With the popularization of vehicles and the development of modern traffic technologies, the license plate recognition technology is widely applied to urban roads, expressways, parking lots, various shopping malls, logistics and other scenes. The license plate is used as important identification information of the vehicle, is very important in vehicle detection and recognition, is an important component of the vehicle detection and recognition, and has important practical significance.
In a traditional license plate recognition method, a vehicle monitoring system is generally used for collecting a license plate image, and then the collected license plate image is subjected to license plate recognition. Due to the influence of factors such as complicated people flow, traffic flow, high speed, poor illumination and the like, the quality of the acquired license plate image is very different, and the conditions of shielding, light reflection, motion blur and the like exist.
Disclosure of Invention
Based on the above technical problem, a license plate image screening model construction method, a license plate image screening model construction device, a storage medium, a computer device, a license plate image screening method, a license plate image screening device, a storage medium and a computer device capable of supporting license plate recognition error reduction are provided.
A construction method of a license plate image screening model comprises the following steps:
acquiring error distribution sample data of a license plate image;
carrying out feature extraction on error distribution sample data of the license plate image, and carrying out model training on a preset identification network through the extracted features;
and adjusting the cross entropy loss function of the identification network according to the model training result until the cross entropy loss function is minimized to obtain a license plate image screening model.
An apparatus for constructing a license plate image screening model, the apparatus comprising:
the license plate image processing device comprises a sample acquisition module, a license plate image processing module and a license plate image processing module, wherein the sample acquisition module is used for acquiring error distribution sample data of a license plate image, and the error distribution sample data comprises license plate character recognition correct data and license plate character recognition error data;
the model training module is used for carrying out model training on a preset identification network based on the license plate character recognition correct data and the license plate character recognition error data;
and the model generation module is used for adjusting the parameters of the identification network according to the model training result until a preset stop condition is met, so as to obtain a license plate image screening model.
A computer-readable storage medium, in which a computer program is stored which, when executed by a processor, causes the processor to carry out the steps of the above-mentioned method.
A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the above method.
According to the construction method, the device, the storage medium and the computer equipment of the license plate image screening model, the error distribution sample data of the license plate image is obtained, the feature extraction is carried out on the error distribution sample data of the license plate image, the model training is carried out on the preset identification network according to the extracted feature, the cross entropy loss function of the identification network is adjusted according to the model training result until the cross entropy loss function is minimized, the license plate image screening model is obtained, the error distribution of the license plate image is learned through the error distribution sample data of the license plate image, subsequently, when the license plate is identified, the license plate image is screened through the license plate image screening model, abnormal data in the license plate image is filtered through a front screening mode, and the high-precision identification of the license plate at the later stage.
A license plate image screening method, comprising:
acquiring a license plate image to be recognized;
reading a preset license plate image screening model, wherein the license plate image screening model is constructed by a construction method of the license plate image screening model;
inputting the license plate image to be recognized into the license plate image screening model;
and screening the license plate image to be identified according to the output result of the license plate image screening model to obtain the screened license plate image.
A license plate image screening device, the device comprising:
the image acquisition module is used for acquiring a license plate image to be identified;
the license plate image screening module is used for screening license plate images, and the license plate images are obtained by a model reading module;
the data import module is used for inputting the license plate image to be recognized into the license plate image screening model;
and the data screening module is used for screening the license plate image to be identified according to the output result of the license plate image screening model to obtain the screened license plate image.
A computer-readable storage medium, in which a computer program is stored which, when executed by a processor, causes the processor to carry out the steps of the above-mentioned method.
A computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of the above method.
According to the license plate image screening method, the license plate image screening device, the storage medium and the computer equipment, the license plate image to be identified is screened through the license plate image screening model to obtain the screened license plate image, and when the license plate image to be identified is subsequently identified based on the screened license plate image, the error of license plate identification can be reduced, and high-precision license plate identification is realized.
Drawings
FIG. 1 is an application environment diagram of a license plate image screening model construction method in one embodiment;
FIG. 2 is a schematic flow chart illustrating a method for constructing a license plate image screening model according to an embodiment;
FIG. 3 is a schematic flow chart of the error distribution sample data acquisition step in one embodiment;
FIG. 4 is a schematic flow chart of a license plate image screening model training process in one embodiment;
FIG. 5 is a schematic flow chart illustrating a license plate image screening method according to an embodiment;
FIG. 6 is a schematic flow chart of license plate image recognition according to an embodiment;
FIG. 7 is a block diagram of an apparatus for constructing a license plate image screening model according to an embodiment;
FIG. 8 is a block diagram of a license plate image screening apparatus according to an embodiment;
FIG. 9 is a block diagram of a computer device in one embodiment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
The construction method of the license plate image screening model provided by the application can be applied to the application environment shown in the figure 1. Wherein a client terminal 102 communicates with a server 104 over a network. The user inputs error distribution sample data of the license plate image through the client terminal 102. The server 104 obtains error distribution sample data of the license plate image, performs feature extraction on the error distribution sample data of the license plate image, performs model training on a preset identification network through the extracted features, adjusts a cross entropy loss function of the identification network according to a model training result until the cross entropy loss function is minimized, and obtains a license plate image screening model. The client terminal 102 may be, but is not limited to, various personal computers, notebook computers, smart phones, and tablet computers, and the server 104 may be implemented by an independent server or a server cluster composed of a plurality of servers.
In an embodiment, as shown in fig. 2, a method for constructing a license plate image screening model is provided, which is described by taking the method as an example for being applied to the server in fig. 1, and includes the following steps:
step 202, obtaining error distribution sample data of the license plate image.
The error distribution sample data of the license plate image refers to error distribution data generated by processing license plate image sample data based on a preset license plate recognition model. The error distribution sample data comprises correct license plate character recognition data and wrong license plate character recognition data. The license plate character recognition correct data refers to data, in sample data, of which the recognition result output by a license plate image through a preset license plate recognition model is consistent with the real condition of the license plate image. For example, a license plate image with an actual label of 'true' is subjected to a preset license plate recognition model to output a label result of 'true'; and the labeling result of the license plate image with the actual label of 'false' output by the preset license plate recognition model is 'false'. The license plate character recognition error data refers to data, in sample data, of which the recognition result output by a preset license plate recognition model of a license plate image is inconsistent with the real condition of the license plate image. For example, a license plate image with an actual label of 'true' is output through a preset license plate recognition model, and a label result of the license plate image with the actual label of 'false' is output through the preset license plate recognition model, and is 'true'.
And step 204, extracting the characteristics of error distribution sample data of the license plate image, and performing model training on a preset identification network through the extracted characteristics.
Inputting a license plate image data set manually marked in advance, such as 20 thousands of license plate image data sets, into a preset license plate recognition model to obtain error distribution data of the data set, wherein the error distribution data are respectively a correctly recognized sample and an incorrectly recognized sample. The identified correct samples and the identified incorrect samples can be obtained according to the recall rate and the accuracy data of the model algorithm. For example, the recall ratio is a% and the precision ratio is B%, where precision is a ratio of correctly retrieved results to all actually retrieved results, and recall is a ratio of correctly retrieved results to all results to be retrieved. And extracting features of the correctly recognized samples and the incorrectly recognized samples, and performing model training on a preset identification network through the extracted features, for example, performing model training on a GoogleLeNet algorithm based on error distribution sample data of the license plate image.
And step 206, adjusting the cross entropy loss function of the identification network according to the model training result until the cross entropy loss function is minimized, and obtaining a license plate image screening model.
Assuming the input is x, the model is calculatedProbability to class k
Assuming a true distribution of q (k), a cross-entropy loss function
Minimizing cross-entropy loss functions, i.e. maximizing likelihood functions, which derive logical outputs
For example, feature extraction is performed on two data blocks, namely a correctly recognized sample and an incorrectly recognized sample, in error distribution sample data of the license plate image through a GoogLeNet algorithm, cross entropy is used for evaluation, and a classification effect is finally achieved, so that a license plate image screening model is generated.
According to the construction method of the license plate image screening model, the error distribution sample data of the license plate image is obtained, the feature extraction is carried out on the error distribution sample data of the license plate image, the model training is carried out on a preset identification network through the extracted feature, the cross entropy loss function of the identification network is adjusted according to the model training result until the cross entropy loss function is minimized, the license plate image screening model is obtained, the error distribution of the license plate image is learned through the error distribution sample data of the license plate image, subsequently, when the license plate is identified, the license plate image is firstly screened through the license plate image screening model, abnormal data in the license plate image is filtered through a front screening mode, and the high-precision identification of the license plate at the later stage is supported.
In one embodiment, as shown in fig. 3, obtaining error distribution sample data of a license plate image includes: step 302, obtaining license plate image sample data; step 304, performing convolution processing on the license plate image sample data, and performing feature vector frame regression processing according to the feature map after the convolution processing to obtain license plate character positioning data; step 306, extracting a statistical characteristic diagram of the license plate character positioning data, and performing nonlinear characteristic extraction processing according to the statistical characteristic diagram to obtain license plate character recognition correct data and license plate character recognition error data; and 308, obtaining error distribution sample data of the license plate image according to the correct license plate character recognition data and the wrong license plate character recognition data. The license plate image sample data refers to a known real license plate character region and a license plate image of real license plate characters. The license plate characters can be positioned through the license plate character positioning model, and the classification of the license plate characters can be realized through the license plate character recognition model. For example, license plate image sample data can be input into a license plate character positioning model, and the license plate character positioning model performs operations such as convolution processing, feature map conversion into feature vectors, feature vector regression processing and the like on the license plate image sample data to obtain license plate character positioning data. And inputting the license plate character positioning data into a license plate character recognition model, and performing operations such as statistical feature extraction, dimension reduction processing, nonlinear feature extraction and the like on the license plate character positioning data by the license plate character recognition model to obtain license plate character recognition correct data and license plate character recognition error data.
In one embodiment, before performing convolution processing on the license plate image sample data and performing regression processing on a feature vector frame according to the feature map after the convolution processing to obtain license plate character positioning data, the method further includes: acquiring license plate image training data, and respectively carrying out normalization processing on each license plate image in the training data to obtain a license plate image after the normalization processing; carrying out convolution operation on the license plate images after the normalization processing through convolution kernels with different preset sizes respectively to obtain a feature map set; converting feature maps in the feature map set into feature vectors, performing frame regression processing based on the feature vectors to determine a license plate character prediction frame, and calculating the distance between the license plate character prediction frame and a license plate character real frame; adjusting a loss function of a preset neural network based on the distance calculation result until the distance calculation result meets a preset condition, and obtaining a license plate character positioning model; carrying out convolution processing on the license plate image sample data, carrying out feature vector frame regression processing according to the feature graph after the convolution processing, and obtaining license plate character positioning data comprises the following steps: and inputting the license plate image sample data to a license plate character positioning model to obtain license plate character positioning data. The normalization process is to normalize a character to an image of the same pixel size in order that the character size may be different when the character is divided. The feature extraction of the license plate image can be realized through a convolution kernel, license plate characters generally consist of thick straight lines or curves, the shape features are obvious, and after the convolution, the visual features such as edges, line segments, corner points and the like of the character image can be detected. The extraction of the features at all positions of the character image was found by convolution of different sizes. A feature graph refers to the corresponding features or characteristics of an object of one class that are distinct from the objects of other classes, or a collection of such features and characteristics. Combining a plurality of or a plurality of characteristics of a certain class of objects together to form a characteristic vector to represent the class of objects, wherein if only a single numerical characteristic exists, the characteristic vector is a one-dimensional vector, and if the characteristic vector is a combination of n characteristics, the characteristic vector is an n-dimensional characteristic vector. For example, a four-dimensional vector (x, y, w, h) is typically used for the window to represent the center coordinates and width and height of the window, respectively. The frame regression refers to finding a mapping relation so that the original window is mapped to obtain a regression window close to the real window, and the frame regression can be realized through translation and scale scaling. The regression processing based on the feature vector can specifically be a mapping relation is found, so that the license plate character original frame constructed based on the feature vector is mapped to obtain a license plate character prediction frame close to the license plate character real frame.
In the license plate image training data, a license plate character region of each license plate image can be calibrated in advance, for example, a pixel of a coordinate point (x, y) belongs to a character, and each license plate image corresponds to a rectangle containing all character pixels in the image. Taking an example of training a SSD (Single Shot Multi Box) target detection algorithm to obtain a license plate character positioning model, performing convolution operation on normalized license plate images through convolution kernels with different sizes to obtain a feature map, converting the feature map into a feature vector, and performing regression processing on the feature vector to obtain an external rectangle. And finally, evaluating the distance between the generated external rectangle and a pre-calibrated rectangle through the Euclidean distance, wherein when the two distances are within a preset proper range, the pixels contained in the external rectangle are character pixels, and the external rectangle is a license plate character area corresponding to the license plate image.
In one embodiment, before extracting a statistical feature map of license plate character positioning data and performing nonlinear feature extraction processing according to the statistical feature map to obtain license plate character recognition correct data and license plate character recognition error data, the method further includes: acquiring license plate character recognition training data, and extracting a statistical characteristic diagram set of a license plate character area in the training data; performing dimension reduction processing on the statistical feature map set, and performing nonlinear feature extraction on the statistical feature map subjected to the dimension reduction processing to obtain a license plate character recognition result; when the license plate character recognition result is inconsistent with the actual license plate character, adjusting a loss function of a preset neural network until the license plate character recognition result is consistent with the actual license plate character to obtain a license plate character recognition model; extracting a statistical characteristic diagram of the license plate character positioning data, and performing nonlinear characteristic extraction processing according to the statistical characteristic diagram to obtain license plate character recognition correct data and license plate character recognition error data, wherein the license plate character recognition correct data and the license plate character recognition error data comprise the following steps: and inputting the license plate character positioning data into a license plate character recognition model to obtain license plate character recognition correct data and license plate character recognition error data. The statistical characteristic graph is obtained by extracting character characteristics in a license plate character region, for example, counting characteristics such as character stroke slope characteristics and character side depth and the like as character extraction characteristics. Too high dimensionality of the feature vectors increases computational complexity, brings adverse effects to subsequent classification problems, and too high dimensionality of the feature vectors negatively affects classification performance, so that dimensionality reduction needs to be performed on the acquired statistical features, and dimensionality reduction can be specifically performed in a feature selection or feature extraction mode. The dimension reduction process enables high-dimensional features to be extracted through the convolutional layer, the nonlinear features comprise the high-dimensional features, and the nonlinear features describe character features from the other dimension, so that the character features are more comprehensive, and character classification is more accurate.
The license plate character recognition training data can be license plate character positioning data obtained by processing a license plate image of known license plate characters through a license plate character positioning model. Taking an example of training an Deep OCR (Optical character recognition) algorithm to obtain a license plate character recognition model, the Deep OCR algorithm is a small neural network including 3 convolutional layers, 2 pooling layers, 2 activation functions, and 1 loss function. Wherein, the convolution layer mainly extracts the edge characteristic of each character; the pooling layer is mainly used for reducing the dimension of the image, so that the extraction of high-dimensional features through the convolution layer becomes possible; the activation function is mainly used for improving the nonlinear fitting capability of the neural network, so that the network adapts to the characteristics of different characters and expresses different characteristics; the loss function is mainly used for evaluating the difference between the character features generated by the neural network and the character features classified in advance, and is an optimization target of the network. By a gradient descending method, the neural network learns the characteristics of each license plate character in the license plate image, and finally the characteristics are evaluated through a loss function, so that the neural network has the capacity of classifying the license plate characters.
In one embodiment, the SSD target detection algorithm and Deep OCR algorithm are used for generating error distribution data of the license plate image, and the GoogLeNet algorithm is used for learning the error distribution data. Performing countermeasure training by using a network A (SSD algorithm + Deep OCR algorithm) and a network B (GoogLeNet algorithm), wherein the network A is a license plate image recognition system capable of recognizing characters of a license plate, and errors of the system are learned through the network B. The license plate characters are positioned by using an SSD algorithm, and then the characters are classified by using a Deep OCR algorithm. As shown in fig. 4, the SSD algorithm + Deep OCR algorithm is applied to 20 ten thousand license plate images labeled manually in advance, and error distribution data of the data set is obtained, including identifying a correct data set and identifying an incorrect data set. And performing feature extraction on the error distribution data of the data set by using a GoogLeNet algorithm, and evaluating by using cross entropy to finally achieve the classification effect of the GoogLeNet algorithm. And applying the GoogLeNet algorithm with the classification effect to the 20 ten thousand license plate images to generate two new data blocks A and B, wherein A is a data set judged to be correctly recognized by the license plate recognition algorithm, and B is a data set judged to be not correctly recognized by the license plate recognition algorithm. And training a license plate character recognition algorithm combined by an SSD algorithm and a Deep OCR algorithm by using a data set A, and then repeatedly performing model training on the license plate character recognition algorithm until the recell of the license plate character recognition algorithm is more than 99 percent and the precision is more than 99 percent.
In one embodiment, as shown in fig. 5, a license plate image screening method is provided, and the method includes: step 502, acquiring a license plate image to be recognized; step 504, reading a preset license plate image screening model, wherein the license plate image screening model is constructed by a construction method of a license plate image screening device; step 506, inputting a license plate image to be recognized into a license plate image screening model; and step 508, screening the license plate image to be recognized according to the output result of the license plate image screening model to obtain the screened license plate image. The license plate image screening method further comprises the steps of reading a preset license plate image recognition model; and inputting the screened license plate image into a license plate image recognition model to obtain a recognition result of the license plate image to be recognized.
According to the license plate image screening method, the license plate image to be recognized is screened through the license plate image screening model to obtain the screened license plate image, and when the license plate image to be recognized is subsequently recognized based on the screened license plate image, the license plate image screening method can reduce the error of license plate recognition and realize high-precision license plate recognition.
In one embodiment, as shown in fig. 6, the license plate image to be recognized is input to the google lenet algorithm with a classification effect, the license plate image which does not accord with the license plate character recognition algorithm is screened out, then the screened license plate image is subjected to license plate character positioning through the SSD algorithm and is subjected to character classification through the Deep OCR algorithm, and a license plate character string result is obtained. Therefore, the license plate image with high quality can be well selected in a complex scene, the character recognition rate of the license plate image is greatly improved, the license plate recognition system can be used for vehicle management under a complex background, and the monitoring and management capability of the vehicle is improved.
It should be understood that although the various steps in the flow charts of fig. 2-6 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2-6 may include multiple sub-steps or multiple stages that are not necessarily performed at the same time, but may be performed at different times, and the order of performance of the sub-steps or stages is not necessarily sequential, but may be performed in turn or alternating with other steps or at least some of the sub-steps or stages of other steps.
Based on the same idea as the above method, fig. 7 is a schematic structural diagram of a license plate image screening model construction apparatus according to an embodiment, where the license plate image screening model construction apparatus includes:
the system comprises a sample acquisition module 702, a license plate recognition module and a license plate recognition module, wherein the sample acquisition module is used for acquiring error distribution sample data of a license plate image, and the error distribution sample data comprises license plate character recognition correct data and license plate character recognition error data;
the model training module 704 is used for performing model training on a preset identification network based on the license plate character recognition correct data and the license plate character recognition error data;
and the model generation module 706 is configured to adjust parameters of the identification network according to a model training result until a preset stop condition is met, so as to obtain a license plate image screening model.
In one embodiment, the sample acquiring module is further configured to acquire sample data of the license plate image; carrying out convolution processing on the license plate image sample data, and carrying out feature vector frame regression processing according to the feature map after the convolution processing to obtain license plate character positioning data; extracting a statistical characteristic diagram of the license plate character positioning data, and performing nonlinear characteristic extraction processing according to the statistical characteristic diagram to obtain license plate character recognition correct data and license plate character recognition error data; and obtaining error distribution sample data of the license plate image according to the correct license plate character recognition data and the wrong license plate character recognition data.
In one embodiment, the device for constructing the license plate image screening model further comprises a license plate character positioning model training module, which is used for acquiring license plate image training data, and respectively carrying out normalization processing on each license plate image in the training data to obtain a license plate image after the normalization processing; carrying out convolution operation on the license plate images after the normalization processing through convolution kernels with different preset sizes respectively to obtain a feature map set; converting feature maps in the feature map set into feature vectors, performing frame regression processing based on the feature vectors to determine a license plate character prediction frame, and calculating the distance between the license plate character prediction frame and a license plate character real frame; adjusting a loss function of a preset neural network based on the distance calculation result until the distance calculation result meets a preset condition, and obtaining a license plate character positioning model; the sample acquisition module is also used for inputting the license plate image sample data into the license plate character positioning model to acquire license plate character positioning data.
In one embodiment, the device for constructing the license plate image screening model further comprises a license plate character recognition model training module, which is used for acquiring license plate character recognition training data and extracting a statistical feature map set of a license plate character region in the training data; performing dimension reduction processing on the statistical feature map set, and performing nonlinear feature extraction on the statistical feature map subjected to the dimension reduction processing to obtain a license plate character recognition result; when the license plate character recognition result is inconsistent with the actual license plate character, adjusting a loss function of a preset neural network until the license plate character recognition result is consistent with the actual license plate character to obtain a license plate character recognition model; the sample acquisition module is also used for inputting the license plate character positioning data into the license plate character recognition model to acquire license plate character recognition correct data and license plate character recognition error data.
Fig. 8 is a schematic structural diagram illustrating a license plate image screening apparatus according to an embodiment, where the license plate image screening apparatus includes:
the image acquisition module 802 is used for acquiring a license plate image to be identified;
the model reading module 804 is used for reading a preset license plate image screening model, and the license plate image screening model is constructed by a construction method of the license plate image screening model;
the data import module 806 is configured to input the license plate image to be recognized to the license plate image screening model;
and the data screening module 808 is configured to screen the license plate image to be recognized according to an output result of the license plate image screening model, so as to obtain a screened license plate image.
In one embodiment, the license plate image screening device further comprises a license plate recognition module, which is used for reading a preset license plate image recognition model; and inputting the screened license plate image into a license plate image recognition model to obtain a recognition result of the license plate image to be recognized.
FIG. 9 is a diagram illustrating an internal structure of a computer device in one embodiment. The computer device may specifically be the server in fig. 1. As shown in fig. 9, the computer apparatus includes a processor, a memory, a network interface, an input device, a display screen, a camera, a sound collection device, and a speaker, which are connected by a system bus. Wherein the memory includes a non-volatile storage medium and an internal memory. The non-volatile storage medium of the computer device stores an operating system and may further store a computer program, and when the computer program is executed by the processor, the computer program may cause the processor to implement a method of constructing a license plate image screening model or a method of screening a license plate image. The internal memory may also store a computer program, and when the computer program is executed by the processor, the processor may execute a method for constructing a license plate image screening model or a method for screening a license plate image. The display screen of the computer equipment can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer equipment can be a touch layer covered on the display screen, a key, a track ball or a touch pad arranged on the shell of the computer equipment, an external keyboard, a touch pad or a mouse and the like.
Those skilled in the art will appreciate that the architecture shown in fig. 9 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.
In one embodiment, the license plate image screening model constructing apparatus or the license plate image screening apparatus provided in the present application may be implemented in the form of a computer program, and the computer program may be executed on a computer device as shown in fig. 9. The memory of the computer device may store the program modules constituting the computer device, and the computer program constituted by the program modules enables the processor to execute the method for constructing the license plate image screening model or the steps of the method for screening the license plate image described in the present specification in the embodiments of the present application.
In one embodiment, a computer device is provided, which includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the method for constructing the license plate image screening model or the steps of the license plate image screening method. The steps of the method for constructing the license plate image screening model or the method for screening the license plate image may be steps of the method for constructing the license plate image screening model or the method for screening the license plate image in each of the embodiments.
In one embodiment, a computer-readable storage medium is provided, which stores a computer program, and when the computer program is executed by a processor, the processor is enabled to execute the method for constructing the license plate image screening model or the steps of the method for screening the license plate image. The steps of the method for constructing the license plate image screening model or the method for screening the license plate image may be steps of the method for constructing the license plate image screening model or the method for screening the license plate image in each of the embodiments.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by a computer program, which can be stored in a non-volatile computer-readable storage medium, and can include the processes of the embodiments of the methods described above when the program is executed. Any reference to memory, storage, database, or other medium used in the embodiments provided herein may include non-volatile and/or volatile memory, among others. Non-volatile memory can include read-only memory (ROM), Programmable ROM (PROM), Electrically Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDRSDRAM), Enhanced SDRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), Rambus Direct RAM (RDRAM), direct bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the present application. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.
Claims (10)
1. A construction method of a license plate image screening model is characterized by comprising the following steps:
acquiring error distribution sample data of a license plate image;
carrying out feature extraction on error distribution sample data of the license plate image, and carrying out model training on a preset identification network through the extracted features;
and adjusting the cross entropy loss function of the identification network according to the model training result until the cross entropy loss function is minimized to obtain a license plate image screening model.
2. The method of claim 1, wherein the obtaining error distribution sample data of the license plate image comprises:
acquiring license plate image sample data;
carrying out convolution processing on the license plate image sample data, and carrying out feature vector frame regression processing according to the feature map after the convolution processing to obtain license plate character positioning data;
extracting a statistical characteristic diagram of the license plate character positioning data, and performing nonlinear characteristic extraction processing according to the statistical characteristic diagram to obtain license plate character recognition correct data and license plate character recognition error data;
and obtaining error distribution sample data of the license plate image according to the license plate character recognition correct data and the license plate character recognition error data.
3. The method according to claim 2, wherein before performing convolution processing on the license plate image sample data and performing feature vector border regression processing according to the feature map after the convolution processing to obtain license plate character positioning data, the method further comprises:
acquiring license plate image training data, and respectively carrying out normalization processing on each license plate image in the training data to obtain a license plate image after the normalization processing;
carrying out convolution operation on the license plate images after the normalization processing through convolution kernels with different preset sizes respectively to obtain a feature map set;
converting the feature maps in the feature map set into feature vectors, performing frame regression processing based on the feature vectors to determine a license plate character prediction frame, and calculating the distance between the license plate character prediction frame and a license plate character real frame;
adjusting a loss function of a preset neural network based on the distance calculation result until the distance calculation result meets a preset condition, and obtaining a license plate character positioning model;
the convolution processing is carried out on the license plate image sample data, and the regression processing of the feature vector frame is carried out according to the feature map after the convolution processing, so that the license plate character positioning data is obtained, wherein the convolution processing comprises the following steps:
and inputting the license plate image sample data to the license plate character positioning model to obtain license plate character positioning data.
4. The method of claim 2, wherein before the extracting the statistical feature map of the license plate character positioning data and performing the non-linear feature extraction processing according to the statistical feature map to obtain the license plate character recognition correct data and the license plate character recognition error data, the method further comprises:
acquiring license plate character recognition training data, and extracting a statistical characteristic diagram set of a license plate character area in the training data;
performing dimension reduction processing on the statistical feature map set, and performing nonlinear feature extraction on the statistical feature map set subjected to the dimension reduction processing to obtain a license plate character recognition result;
when the license plate character recognition result is inconsistent with the actual license plate character, adjusting a loss function of a preset neural network until the license plate character recognition result is consistent with the actual license plate character to obtain a license plate character recognition model;
the step of extracting the statistical characteristic diagram of the license plate character positioning data, and performing nonlinear characteristic extraction processing according to the statistical characteristic diagram to obtain license plate character recognition correct data and license plate character recognition error data comprises the following steps:
and inputting the license plate character positioning data into the license plate character recognition model to obtain license plate character recognition correct data and license plate character recognition error data.
5. A license plate image screening method, comprising:
acquiring a license plate image to be recognized;
reading a preset license plate image screening model, wherein the license plate image screening model is constructed by the method of any one of claims 1-4;
inputting the license plate image to be recognized into the license plate image screening model;
and screening the license plate image to be identified according to the output result of the license plate image screening model to obtain the screened license plate image.
6. The method of claim 5, further comprising:
reading a preset license plate image recognition model;
and inputting the screened license plate image into the license plate image recognition model to obtain a recognition result of the license plate image to be recognized.
7. A construction device of a license plate image screening model is characterized by comprising:
the license plate image processing device comprises a sample acquisition module, a license plate image processing module and a license plate image processing module, wherein the sample acquisition module is used for acquiring error distribution sample data of a license plate image, and the error distribution sample data comprises license plate character recognition correct data and license plate character recognition error data;
the model training module is used for carrying out model training on a preset identification network based on the license plate character recognition correct data and the license plate character recognition error data;
and the model generation module is used for adjusting the parameters of the identification network according to the model training result until a preset stop condition is met, so as to obtain a license plate image screening model.
8. A license plate image screening device, characterized in that the device includes:
the image acquisition module is used for acquiring a license plate image to be identified;
the model reading module is used for reading a preset license plate image screening model, and the license plate image screening model is constructed by the method of any one of claims 1 to 4;
the data import module is used for inputting the license plate image to be recognized into the license plate image screening model;
and the data screening module is used for screening the license plate image to be identified according to the output result of the license plate image screening model to obtain the screened license plate image.
9. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, causes the processor to carry out the steps of the method according to any one of claims 1 to 6.
10. A computer device comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method according to any one of claims 1 to 6.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911239521.1A CN112926610A (en) | 2019-12-06 | 2019-12-06 | Construction method of license plate image screening model and license plate image screening method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911239521.1A CN112926610A (en) | 2019-12-06 | 2019-12-06 | Construction method of license plate image screening model and license plate image screening method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112926610A true CN112926610A (en) | 2021-06-08 |
Family
ID=76161509
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911239521.1A Pending CN112926610A (en) | 2019-12-06 | 2019-12-06 | Construction method of license plate image screening model and license plate image screening method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112926610A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116935373A (en) * | 2023-09-18 | 2023-10-24 | 珠海亿智电子科技有限公司 | Training method, device, equipment and storage medium for multi-type license plate recognition model |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104809443A (en) * | 2015-05-05 | 2015-07-29 | 上海交通大学 | Convolutional neural network-based license plate detection method and system |
| CN106295646A (en) * | 2016-08-10 | 2017-01-04 | 东方网力科技股份有限公司 | A kind of registration number character dividing method based on degree of depth study and device |
| EP3182334A1 (en) * | 2015-12-17 | 2017-06-21 | Xerox Corporation | License plate recognition using coarse-to-fine cascade adaptations of convolutional neural networks |
| CN108830294A (en) * | 2018-05-09 | 2018-11-16 | 四川斐讯信息技术有限公司 | A kind of augmentation method of image data |
| CN109658401A (en) * | 2018-12-14 | 2019-04-19 | 上海商汤智能科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN109753914A (en) * | 2018-12-28 | 2019-05-14 | 安徽清新互联信息科技有限公司 | A kind of license plate character recognition method based on deep learning |
| CN110298394A (en) * | 2019-06-18 | 2019-10-01 | 中国平安财产保险股份有限公司 | A kind of image-recognizing method and relevant apparatus |
| KR102030628B1 (en) * | 2019-04-04 | 2019-10-10 | (주)아이엠시티 | Recognizing method and system of vehicle license plate based convolutional neural network |
-
2019
- 2019-12-06 CN CN201911239521.1A patent/CN112926610A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104809443A (en) * | 2015-05-05 | 2015-07-29 | 上海交通大学 | Convolutional neural network-based license plate detection method and system |
| EP3182334A1 (en) * | 2015-12-17 | 2017-06-21 | Xerox Corporation | License plate recognition using coarse-to-fine cascade adaptations of convolutional neural networks |
| CN106295646A (en) * | 2016-08-10 | 2017-01-04 | 东方网力科技股份有限公司 | A kind of registration number character dividing method based on degree of depth study and device |
| CN108830294A (en) * | 2018-05-09 | 2018-11-16 | 四川斐讯信息技术有限公司 | A kind of augmentation method of image data |
| CN109658401A (en) * | 2018-12-14 | 2019-04-19 | 上海商汤智能科技有限公司 | Image processing method and device, electronic equipment and storage medium |
| CN109753914A (en) * | 2018-12-28 | 2019-05-14 | 安徽清新互联信息科技有限公司 | A kind of license plate character recognition method based on deep learning |
| KR102030628B1 (en) * | 2019-04-04 | 2019-10-10 | (주)아이엠시티 | Recognizing method and system of vehicle license plate based convolutional neural network |
| CN110298394A (en) * | 2019-06-18 | 2019-10-01 | 中国平安财产保险股份有限公司 | A kind of image-recognizing method and relevant apparatus |
Non-Patent Citations (2)
| Title |
|---|
| FAHIMEH RAMAZANKHANI; MAHDI YAZDIAN-DEHKORDI: "Iranian License Plate Detection using Cascade Classifier", 2019 27TH IRANIAN CONFERENCE ON ELECTRICAL ENGINEERING (ICEE), 30 April 2019 (2019-04-30), pages 1860 - 1863, XP033588889, DOI: 10.1109/IranianCEE.2019.8786468 * |
| PHAM, VAN HUY AND DINH, PHONG QUANG AND NGUYEN, VAN HUAN: "CNN-based character recognition for license plate recognition system", SPRINGER, 14 February 2018 (2018-02-14), pages 594, XP047465143, DOI: 10.1007/978-3-319-75420-8_56 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116935373A (en) * | 2023-09-18 | 2023-10-24 | 珠海亿智电子科技有限公司 | Training method, device, equipment and storage medium for multi-type license plate recognition model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109492643B (en) | Certificate identification method and device based on OCR, computer equipment and storage medium | |
| CN110414507B (en) | License plate recognition method and device, computer equipment and storage medium | |
| US11657602B2 (en) | Font identification from imagery | |
| WO2019232853A1 (en) | Chinese model training method, chinese image recognition method, device, apparatus and medium | |
| CN110569721A (en) | Recognition model training method, image recognition method, device, equipment and medium | |
| CN111523414B (en) | Face recognition method, device, computer equipment and storage medium | |
| CN109711416B (en) | Target identification method and device, computer equipment and storage medium | |
| Zhang et al. | Road recognition from remote sensing imagery using incremental learning | |
| WO2019232850A1 (en) | Method and apparatus for recognizing handwritten chinese character image, computer device, and storage medium | |
| CN113837151B (en) | Table image processing method and device, computer equipment and readable storage medium | |
| CN105335760A (en) | Image number character recognition method | |
| WO2022134354A1 (en) | Vehicle loss detection model training method and apparatus, vehicle loss detection method and apparatus, and device and medium | |
| CN114092938B (en) | Image recognition processing method and device, electronic equipment and storage medium | |
| CN111507337A (en) | License plate recognition method based on hybrid neural network | |
| Yépez et al. | Deep learning‐based embedded license plate localisation system | |
| CN112861970A (en) | Fine-grained image classification method based on feature fusion | |
| CN115496976A (en) | Visual processing method, device, equipment and medium for multi-source heterogeneous data fusion | |
| CN110580507B (en) | City texture classification and identification method | |
| CN111368632A (en) | Signature identification method and device | |
| CN110796145A (en) | Multi-certificate segmentation association method based on intelligent decision and related equipment | |
| CN112926610A (en) | Construction method of license plate image screening model and license plate image screening method | |
| CN110163910B (en) | Object positioning method, device, computer equipment and storage medium | |
| CN111582057A (en) | Face verification method based on local receptive field | |
| CN111488811A (en) | Face recognition method and device, terminal equipment and computer readable medium | |
| CN115984588A (en) | Image background similarity analysis method and device, electronic equipment and storage medium |
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 |