CN107563385B - License plate character recognition method based on depth convolution production confrontation network - Google Patents
License plate character recognition method based on depth convolution production confrontation network Download PDFInfo
- Publication number
- CN107563385B CN107563385B CN201710781905.0A CN201710781905A CN107563385B CN 107563385 B CN107563385 B CN 107563385B CN 201710781905 A CN201710781905 A CN 201710781905A CN 107563385 B CN107563385 B CN 107563385B
- Authority
- CN
- China
- Prior art keywords
- license plate
- convolutional layer
- model
- network
- picture
- 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.)
- Active
Links
Abstract
The invention proposes a kind of license plate character recognition methods based on depth convolution production confrontation network, and the specific implementation steps are as follows: (1) extracting license plate picture to be identified；(2) it constructs and training depth convolution production fights network DCGAN；(3) license plate picture is generated；(4) sample set of character recognition network is constructed；(5) it constructs and trains Recognition of License Plate Characters network C NN；(6) Recognition of License Plate Characters.The present invention uses the license plate character recognition method based on depth convolution production confrontation network, can effectively overcome in the prior art license plate data it is seriously deficient, training sample leads to the shortcomings that network over-fitting less, effectively enhance data sample, and make the generalization ability of character recognition network and robustness stronger, improve Recognition of License Plate Characters rate.
Description
Technical field
The invention belongs to technical field of image processing, and it is raw based on depth convolution to further relate to one of deep learning
The character identifying method of accepted way of doing sth confrontation network.The present invention is for obtained by the high definition photographing device in traffic system, being arranged from crossing
To picture in extract license plate image, and then a large amount of license plate image is generated with a small amount of license plate image, by the license plate image of generation
Processing is training sample, and training Recognition of License Plate Characters network realizes Recognition of License Plate Characters.
Background technique
With the continuous improvement of social and economic level and popularizing for vehicle, the ever-expanding communication of scale is to more intelligence
The demand of the technology and systems of energyization is bigger, and intelligent transportation system has become the hot issue of social life.Vehicle identification system
The important component united as intelligent transportation system, remembers automatically in expressway access, parking lot unattended, violation vehicle
The fields such as record have this to be widely applied, and important composition of the Recognition of License Plate Characters as vehicle identification system, its realization has
Very big economic value and realistic meaning.
With the raising of computer hardware capacity and the speed of service, deep learning algorithm can carry out image real-time
Processing, so having there is more and more deep learning algorithms to apply in Car license recognition.It applies in Car license recognition at present
Deep learning method such as BP neural network, CNN etc., however deep learning needs a large amount of labeled data as training sample,
And often only have partial data can be used in the data of magnanimity in reality, and domestic license plate recognition technology research is existing
Shape is business and privatization, and researcher is difficult to find the research that open data carry out license plate recognition technology, in data trade net
It stands, license plate data are also fairly expensive, so the rare of license plate data is asked in the license plate recognition technology of application deep learning
Inscribe the deep application for influencing deep learning method.
A kind of patent document " license plate character recognition method " of the Jiuzhou Electrical Appliance Group Co., Ltd., Sichuan in its application
A kind of license plate character recognition method is proposed in (number of patent application: 201210587347.1, publication number: CN103065137B).
This method use first canny algorithm and bianry image in conjunction with method identify character, when identification, has extracted the side of character
Edge and jump information, then the jump of character edge pixel is matched with the jump of template character set, it is highest to find matching degree
Template character, and then obtain the recognition result of character.This character recognition based on jump, preferably solves under various interference
Character recognition problem keeps relatively stable high discrimination.But the shortcoming that this method still has is recognition speed
Relatively slow, discrimination and recognition speed are difficult to meet simultaneously, and the vehicle identification for part without license plate or license plate serious damage
Rate be it is rather low, real-time in practical application, accuracy requirement cannot be met very well.
Dong Jun prince wife, in the paper " Recognition of License Plate Characters based on convolutional neural networks " that Zheng Baichuan, Yang Zejing are delivered at it
Propose a kind of license plate character recognition method based on convolutional neural networks.This method carries out size to characters on license plate image first
The pretreatments such as normalization, denoising, binaryzation, refinement, character zone be placed in the middle, remove complex background, obtain simple character shape
Then structure utilizes proposed CNN model to be trained, identify to pretreated characters on license plate collection.The experimental results showed that
This method can reach higher correct recognition rata.But the shortcoming that this method still has is, first, being based on depth
The serious deficient problem of license plate data in the Vehicle License Plate Recognition System of study.Second, needing a large amount of labeled data as training sample
This.Third, training sample is very few to will lead to network over-fitting.
Summary of the invention
It is a kind of based on depth convolution production the purpose of the present invention is in view of the deficiency of the prior art, proposing
The license plate character recognition method of network is fought, present invention sample compared with other technology of vehicle license plate character identification in the prior art is rich
Richness, recognition accuracy is high, and speed is fast, adaptable.
Realizing the thinking of the object of the invention is: it first constructs and training depth convolution production fights network, it will be to be identified
License plate picture fights network by trained depth convolution production and generates a large amount of license plate picture, then by the license plate figure of generation
Piece denoised respectively, binaryzation and Character segmentation, will divide obtained number and letter is built into character recognition network
Then sample set constructs and trains Recognition of License Plate Characters network C NN, test alphabetic collection is finally inputted trained license plate word
Symbol identification network C NN classifies, and obtains final character identification result.
Realize that specific step is as follows for the object of the invention:
(1) license plate picture to be identified is extracted:
Using license plate picture extracting method, from the obtained picture of high definition photographing device, being extracted wait know in traffic intersection
Other license plate picture；
(2) it constructs and training depth convolution production fights network DCGAN:
(2a) building is containing the depth convolutional neural networks being of five storeys as generation model；
(2b) building is containing the convolutional neural networks being of five storeys as discrimination model；
(2c) by trained generation mould and sentences using individually alternately training method, training generate model and discrimination model
The depth convolution production that other model composition generates license plate picture fights network DCGAN；
(3) license plate picture is generated:
License plate photo resolution to be identified is normalized to 256 × 64 pixel sizes by (3a), the license plate that obtains that treated
Picture；
(3b) will treated license plate picture, be input to the depth convolution production confrontation network for generating license plate picture
DCGAN exports the license plate picture of generation；
(4) sample set of character recognition network is constructed:
(4a) from the obtained picture of high definition photographing device, extracted in traffic intersection license plate picture to be identified respectively into
Row denoising, binaryzation and Character segmentation, obtain 7 character pictures, remove chinese character, retain letter and number character, constitute
Training data sample set A；
(4b) obtains 700 figures from the license plate picture of generation, is denoised respectively to 700 figures, binaryzation and word
Symbol segmentation retains letter and number character and constitutes set of data samples B；
(4c) chooses 5% sample after mixing set of data samples A and set of data samples B, remaining as test sample collection
95% sample is as training sample set C；
(5) it constructs and trains Recognition of License Plate Characters network C NN:
(5a) constructs the convolutional neural networks CNN containing 7 layers；
Training sample set C is input in convolutional neural networks CNN by (5b), training convolutional neural networks CNN, until its is defeated
The loss function value of layer is less than or equal to 0.0001 out, obtains trained Recognition of License Plate Characters network C NN；
(6) Recognition of License Plate Characters:
Test sample collection is input in trained Recognition of License Plate Characters network C NN, the alphabetical sum number identified is exported
Word character.
Compared with the prior art, the present invention has the following advantages:
First, since the present invention is using the method constructed and training depth convolution production fights network DCGAN, generate vehicle
Board picture enriches the diversity and randomness of training sample set, enhances the training sample set in character recognition network.Overcome
License plate data are seriously deficient in the Vehicle License Plate Recognition System based on deep learning in the prior art, and training sample is very few will lead to net
Network over-fitting needs the problems such as a large amount of labeled data is as training sample, so that making simultaneously the invention can avoid over-fitting
The generalization ability and robustness of network are stronger, it is only necessary to which low volume data saves the consumption of artificial labeled data as training sample
When.
Second, due to the method that the present invention uses Recognition of License Plate Characters, assembled for training using the sample of building character recognition network
Practice convolutional neural networks CNN and do character recognition, it is slower to overcome recognition speed in existing conventional identification techniques, discrimination and identification
Speed is difficult to the problems such as meeting simultaneously, so that the present invention identifies the fast speed of letter with numerical character, recognition correct rate is high.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Specific embodiment
1 the present invention will be further described with reference to the accompanying drawing.
Step 1, license plate picture to be identified is extracted.
Using license plate picture extracting method, from the obtained picture of high definition photographing device, being extracted wait know in traffic intersection
Other license plate picture, license plate picture extracting method refer to, the obtained picture of high definition photographing device in traffic intersection is used and is cut
Figure tool intercepts 256 × 64 area sizes, the license plate figure to be identified comprising high-visible license plate number and alphabetic character
Piece.
Step 2, it constructs and training depth convolution production fights network DCGAN.
Building is set containing the depth convolutional neural networks being of five storeys as model, this 5 layers of depth convolutional neural networks are generated
Setting is, is from left to right followed successively by full articulamentum, micro-stepping width convolutional layer, convolutional layer, micro-stepping width convolutional layer, convolutional layer.
Containing the convolutional neural networks that are of five storeys as discrimination model, the setting of this 5 layers of convolutional neural networks is for building, from
It is left-to-right to be followed successively by convolutional layer, stride convolutional layer, convolutional layer, stride convolutional layer, full articulamentum.
Using independent alternately training method, training generates model and discrimination model, by trained generation mould and differentiates mould
The depth convolution production that type composition generates license plate picture fights network DCGAN；
Specific step is as follows for the independent alternating training method:
Step 1, fixed discrimination model, training generate model, 1 100 dimension random Gaussian are input to generation model
Full articulamentum obtains the full articulamentum output characteristic pattern of generation model of 512 16 × 4 pixel sizes；
Step 2 will generate the full articulamentum output characteristic pattern of model and be input to generation model micro-stepping width convolutional layer, successively to it
Batch standardization and linear R elu transformation, deconvolution operation are carried out, the generation model micro-stepping of 256 32 × 8 pixel sizes is obtained
Width convolutional layer exports characteristic pattern；
Step 3, by generate model micro-stepping width convolutional layer output characteristic pattern be input to generate model convolutional layer, to its successively into
Row batch standardization and linear R elu transformation, deconvolution operation, obtain the generation model convolutional layer of 128 64 × 16 pixel sizes
Export characteristic pattern；
Step 4, by generate model convolutional layer output characteristic pattern be input to generate model micro-stepping width convolutional layer, to its successively into
Row batch standardization and linear R elu transformation, deconvolution operation, obtain the generation model micro-stepping width of 64 128 × 32 pixel sizes
Convolutional layer exports characteristic pattern；
Step 5, by generate model micro-stepping width convolutional layer output characteristic pattern be input to generate model convolutional layer, to its successively into
Row batch standardization and linear R elu transformation, deconvolution operation, obtain the generation license plate picture of 256 × 64 pixel sizes；
Step 6, it is fixed to generate model, training discrimination model, by the generation license plate picture of 256 × 64 pixel sizes and wait know
Other license plate picture is input to discrimination model convolutional layer jointly, carries out convolution operation to it, exports as 128 × 32 pixel sizes
Discrimination model convolutional layer exports characteristic pattern；
Discrimination model convolutional layer output characteristic pattern is input to discrimination model stride convolutional layer, successively carried out to it by step 7
Standardization, convolution sum linear R elu transformation are criticized, the discrimination model stride convolutional layer of 64 × 16 pixel sizes of output exports feature
Figure；
Discrimination model stride convolutional layer output characteristic pattern is input to discrimination model convolutional layer, successively carried out to it by step 8
Standardization, convolution sum linear R elu transformation are criticized, the discrimination model convolutional layer of 32 × 8 pixel sizes of output exports characteristic pattern；
Discrimination model convolutional layer output characteristic pattern is input to discrimination model stride convolutional layer, successively carried out to it by step 9
Standardization, convolution sum linear R elu transformation are criticized, the discrimination model stride convolutional layer of 16 × 4 pixel sizes of output exports feature
Figure；
Discrimination model stride convolutional layer output characteristic pattern is input to the full articulamentum of discrimination model, carried out to it by step 10
Nonlinear transformation obtains the full articulamentum output probability of discrimination model；
The full articulamentum output probability of discrimination model is transmitted to generation model by step 11, is optimized using optimizer and is differentiated mould
Type and generation model, obtain loss function value；
Step 12, judge loss function value whether and meanwhile meet generate model output layer loss function value be less than or equal to
10.000 and discrimination model output layer loss function value be less than or equal to 0.0001, if so, execute step 13, otherwise, execute
Step 1；
Step 13 obtains trained depth convolution production confrontation network DCGAN.
Step 3, license plate picture is generated.
License plate photo resolution to be identified is normalized to 256 × 64 pixel sizes, the license plate picture that obtains that treated.
Again will treated license plate picture, be input to the depth convolution production confrontation network DCGAN for generating license plate picture, output generates
License plate picture.
Step 4, the sample set of character recognition network is constructed.
It is gone respectively from license plate picture to be identified in the obtained picture of high definition photographing device, is extracted in traffic intersection
It makes an uproar, binaryzation and Character segmentation, obtains 7 character pictures, remove chinese character, retain letter and number character, composing training
Set of data samples A.
700 figures are obtained from the license plate picture of generation, 700 figures are denoised respectively, binaryzation and character point
It cuts, retains letter and number character and constitute set of data samples B.5% is chosen after set of data samples A and set of data samples B is mixed
Sample, as test sample collection, the sample of residue 95% is as training sample set C.
Step 5, it constructs and trains Recognition of License Plate Characters network C NN.
Construct the convolutional neural networks CNN containing 7 layers.The setting of 7 layers of convolutional neural networks CNN is, from left to right according to
It is secondary be convolutional layer Conv1, pond layer Pool2, convolutional layer Conv3, pond layer Pool4, full articulamentum Fc5, full articulamentum Fc6,
Classify layer Softmax7.Training sample set C is input to convolutional neural networks CNN, training convolutional neural networks CNN is defeated until its
The loss function of layer is less than or equal to 0.0001 out, obtains trained Recognition of License Plate Characters network C NN.
Step 6, Recognition of License Plate Characters.
Test sample collection is input in trained Recognition of License Plate Characters network C NN, the alphabetical sum number identified is exported
Word character.
Effect of the invention is further described below with reference to emulation experiment.
1. emulation experiment condition:
The hardware environment that the present invention emulates is: Intel Core (TM) i7-6700@3.40GHZ × 8, GPUNVIDIA
GeForce GTX TITAN X, 8GB memory；Software environment: ubuntu 14.04, Ipython2.7；Windows 7, Matlab
R2015b。
2. emulation content and result:
License plate picture to be identified is input to depth convolution production confrontation net DCGAN first by the present invention, generates 700
License plate picture.Training sample set C is constructed again, and with training sample set C training Recognition of License Plate Characters network C NN, test set is inputted
Classify in trained Recognition of License Plate Characters network C NN classifier, show that the final accuracy rate of Recognition of License Plate Characters is
94.66%.Table 1 is to be input to test set in trained Recognition of License Plate Characters network C NN to identify, after iteration 50 times, statistics
The correct quantity of each character recognition in 34 characters on license plate, discrimination are the correct character quantity of identification and single character
The ratio between total quantity.
Table 1, Recognition of License Plate Characters rate table
| Characters on license plate | 0 | 1 | 2 | 3 | 4 | 5 | 6 |
| Discrimination | 96.5% | 95.42% | 93.67% | 94.18% | 92.42% | 93.47% | 95.42% |
| Characters on license plate | 7 | 8 | 9 | A | B | C | D |
| Discrimination | 96.52% | 93.5% | 95.74% | 96.41% | 91.75% | 97.5% | 91.8% |
| Characters on license plate | E | F | G | H | J | K | L |
| Discrimination | 96.25% | 96.72% | 93.2% | 96.72% | 94.3% | 95.27% | 95.42% |
| Characters on license plate | M | N | P | Q | R | S | T |
| Discrimination | 94.1% | 95.18% | 95.42% | 93.42% | 92.25% | 91.32% | 95.67% |
| Characters on license plate | U | V | W | X | Y | Z | |
| Discrimination | 93.36% | 93.12% | 95.42% | 96.27% | 93.83% | 92.9% |
The table of comparisons 1 includes two in table, and characters on license plate and discrimination, characters on license plate have 34, is followed successively by 0,1,2,3,
4,5,6,7,8,9, A, B, C, D, E, F, G, H, J, K, L, M, N, P, Q, R, S, T, U, V, W, X, Y, Z, the discrimination of each character
Successively are as follows: 96.5%, 95.42%, 93.67%, 94.18%, 92.42%, 93.47%, 95.42%, 96.52%, 93.5%,
95.74%, 96.41%, 91.75%, 97.5%, 91.8%, 96.25%, 96.72%, 93.2%, 96.72%, 94.3%,
95.27%, 95.42%, 94.1%, 95.18%, 95.42%, 93.42%, 92.25%, 91.32%, 95.67%,
93.36%, 93.12%, 95.42%, 96.27%, 93.83%, 92.9%.It can be seen that numerical character 0,1,6,7,9 and word
The discrimination of alphabetic character A, C, E, F, H, K, L, N, P, T, W, X are higher, i.e., the present invention is with higher when identifying characters on license plate
Discrimination.
Claims (6)
1. a kind of license plate character recognition method based on depth convolution production confrontation network, which is characterized in that including walking as follows
It is rapid:
(1) license plate picture to be identified is extracted:
Using license plate picture extracting method, from the obtained picture of high definition photographing device, extraction is to be identified in traffic intersection
License plate picture；
(2) it constructs and training depth convolution production fights network DCGAN:
(2a) building is containing the depth convolutional neural networks being of five storeys as generation model；
(2b) building is containing the convolutional neural networks being of five storeys as discrimination model；
(2c) using individually alternating training method, training generates model and discrimination model, by trained generation model and differentiation
The depth convolution production that model composition generates license plate picture fights network DCGAN；
(3) license plate picture is generated:
License plate photo resolution to be identified is normalized to 256 × 64 pixel sizes by (3a), the license plate picture that obtains that treated；
(3b) will treated license plate picture, be input to the depth convolution production confrontation network DCGAN for generating license plate picture, it is defeated
The license plate picture being born；
(4) sample set of character recognition network is constructed:
(4a) is gone respectively from license plate picture to be identified in the obtained picture of high definition photographing device, is extracted in traffic intersection
It makes an uproar, binaryzation and Character segmentation, obtains 7 character pictures, remove chinese character, retain letter and number character, composing training
Set of data samples A；
(4b) obtains 700 figures from the license plate picture of generation, is denoised respectively to 700 figures, binaryzation and character point
It cuts, retains letter and number character and constitute set of data samples B；
(4c) chooses 5% sample after mixing set of data samples A and set of data samples B, as test sample collection, residue 95%
Sample as training sample set C；
(5) it constructs and trains Recognition of License Plate Characters network C NN:
(5a) constructs the convolutional neural networks CNN containing 7 layers；
Training sample set C is input in convolutional Neural net CNN by (5b), training convolutional neural networks CNN, until its output layer
Loss function value is less than or equal to 0.0001, obtains trained Recognition of License Plate Characters network C NN；
(6) Recognition of License Plate Characters:
Test sample collection is input in trained Recognition of License Plate Characters network C NN, the letter and number word identified is exported
Symbol.
2. the license plate character recognition method according to claim 1 based on depth convolution production confrontation network, feature
Be: license plate picture extracting method described in step (1) refers to, by the obtained picture of high definition photographing device in traffic intersection
Using screenshot tool, 256 × 64 area sizes, the vehicle to be identified comprising high-visible license plate number and alphabetic character are intercepted
Board picture.
3. the license plate character recognition method according to claim 1 based on depth convolution production confrontation network, feature
Be: the setting of 5 layers of depth convolutional neural networks described in step (2a) is from left to right to be followed successively by full articulamentum, micro-stepping
Width convolutional layer, convolutional layer, micro-stepping width convolutional layer, convolutional layer.
4. the license plate character recognition method according to claim 1 based on depth convolution production confrontation network, feature
Be: the setting of 5 layers of convolutional neural networks described in step (2b) is from left to right to be followed successively by convolutional layer, stride convolutional layer,
Convolutional layer, stride convolutional layer, full articulamentum.
5. the license plate character recognition method according to claim 1 based on depth convolution production confrontation network, feature
Be: individually replacing training method described in step (2c), specific step is as follows:
Step 1, fixed discrimination model, training generate model, and 1 100 dimension random Gaussian is input to generation model and is connected entirely
Layer is connect, the full articulamentum output characteristic pattern of generation model of 512 16 × 4 pixel sizes is obtained；
Step 2 will generate the full articulamentum output characteristic pattern of model and be input to generation model micro-stepping width convolutional layer, successively carries out to it
Standardization and linear R elu transformation, deconvolution operation are criticized, the generation model micro-stepping width for obtaining 256 32 × 8 pixel sizes is rolled up
Lamination exports characteristic pattern；
Step 3 will generate model micro-stepping width convolutional layer output characteristic pattern and be input to generation model convolutional layer, successively criticize to it
Standardization and linear R elu transformation, deconvolution operation, obtain the generation model convolutional layer output of 128 64 × 16 pixel sizes
Characteristic pattern；
Step 4 will generate model convolutional layer output characteristic pattern and be input to generation model micro-stepping width convolutional layer, successively criticize to it
Standardization and linear R elu transformation, deconvolution operation, obtain the generation model micro-stepping width convolution of 64 128 × 32 pixel sizes
Layer output characteristic pattern；
Step 5 will generate model micro-stepping width convolutional layer output characteristic pattern and be input to generation model convolutional layer, successively criticize to it
Standardization and linear R elu transformation, deconvolution operation, obtain the generation license plate picture of 256 × 64 pixel sizes；
Step 6, fixed to generate model, training discrimination model, by the generation license plate picture of 256 × 64 pixel sizes and to be identified
License plate picture is input to discrimination model convolutional layer jointly, carries out convolution operation to it, exports as the differentiation of 128 × 32 pixel sizes
Model convolutional layer exports characteristic pattern；
Discrimination model convolutional layer output characteristic pattern is input to discrimination model stride convolutional layer, batch mark is successively carried out to it by step 7
Standardization, convolution sum linear R elu transformation, the discrimination model stride convolutional layer of 64 × 16 pixel sizes of output export characteristic pattern；
Discrimination model stride convolutional layer output characteristic pattern is input to discrimination model convolutional layer, batch mark is successively carried out to it by step 8
Standardization, convolution sum linear R elu transformation, the discrimination model convolutional layer of 32 × 8 pixel sizes of output export characteristic pattern；
Discrimination model convolutional layer output characteristic pattern is input to discrimination model stride convolutional layer, batch mark is successively carried out to it by step 9
Standardization, convolution sum linear R elu transformation, the discrimination model stride convolutional layer of 16 × 4 pixel sizes of output export characteristic pattern；
Discrimination model stride convolutional layer output characteristic pattern is input to the full articulamentum of discrimination model, carried out to it non-thread by step 10
Property transformation, obtain the full articulamentum output probability of discrimination model；
The full articulamentum output probability of discrimination model is transmitted to generation model by step 11, using optimizer optimization discrimination model and
Model is generated, loss function value is obtained；
Step 12, judge loss function value whether and meanwhile meet generate model output layer loss function value be less than or equal to 10.000
It is less than or equal to 0.0001 with the loss function value of discrimination model output layer, if so, executing step 13, otherwise, executes step 1；
Step 13 obtains trained depth convolution production confrontation network DCGAN.
6. the license plate character recognition method according to claim 1 based on depth convolution production confrontation network, feature
Be: the setting of 7 layers of convolutional neural networks CNN described in step (5a) is from left to right to be followed successively by convolutional layer Conv1, pond
Change layer Pool2, convolutional layer Conv3, pond layer Pool4, full articulamentum Fc5, full articulamentum Fc6, classify layer Softmax7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710781905.0A CN107563385B (en) | 2017-09-02 | 2017-09-02 | License plate character recognition method based on depth convolution production confrontation network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710781905.0A CN107563385B (en) | 2017-09-02 | 2017-09-02 | License plate character recognition method based on depth convolution production confrontation network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107563385A CN107563385A (en) | 2018-01-09 |
| CN107563385B true CN107563385B (en) | 2019-10-25 |
Family
ID=60978829
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710781905.0A Active CN107563385B (en) | 2017-09-02 | 2017-09-02 | License plate character recognition method based on depth convolution production confrontation network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107563385B (en) |
Families Citing this family (44)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108229130B (en) * | 2018-01-30 | 2021-04-16 | 中国银联股份有限公司 | Verification method and device |
| CN108333183B (en) * | 2018-01-31 | 2021-03-16 | 西安工程大学 | Yarn-dyed shirt cut piece defect detection method based on DCGAN and DCNN |
| CN108229434A (en) * | 2018-02-01 | 2018-06-29 | 福州大学 | A kind of vehicle identification and the method for careful reconstruct |
| CN108491837B (en) * | 2018-03-07 | 2021-12-17 | 浙江工业大学 | Anti-attack method for improving license plate attack robustness |
| CN108446700B (en) * | 2018-03-07 | 2021-10-29 | 浙江工业大学 | License plate attack generation method based on anti-attack |
| CN108802041B (en) * | 2018-03-16 | 2023-05-05 | 惠州旭鑫智能技术有限公司 | Method for rapidly changing small sample set of screen detection |
| CN108710892B (en) * | 2018-04-04 | 2020-09-01 | 浙江工业大学 | Cooperative immune defense method for multiple anti-picture attacks |
| CN108647683B (en) * | 2018-04-28 | 2021-08-17 | 浙江大学 | Method and system for generating character confrontation verification code based on frequency domain noise addition |
| CN109190648B (en) * | 2018-06-26 | 2020-12-29 | Oppo（重庆）智能科技有限公司 | Simulation environment generation method and device, mobile terminal and computer readable storage medium |
| CN109117863B (en) * | 2018-07-04 | 2021-08-13 | 全球能源互联网研究院有限公司 | Insulator sample expansion method and device based on deep convolution generation countermeasure network |
| CN110689001B (en) * | 2018-07-05 | 2023-06-20 | 山东华软金盾软件股份有限公司 | Method for generating license plate training sample in complex environment |
| CN110728626A (en) * | 2018-07-16 | 2020-01-24 | 宁波舜宇光电信息有限公司 | Image deblurring method and apparatus and training thereof |
| CN109190504B (en) | 2018-08-10 | 2020-12-22 | 百度在线网络技术（北京）有限公司 | Automobile image data processing method and device and readable storage medium |
| CN109143353B (en) * | 2018-08-20 | 2019-10-01 | 电子科技大学 | A kind of pre-stack seismic waveform classification generating confrontation network based on depth convolution |
| CN109447064B (en) * | 2018-10-09 | 2019-07-30 | 温州大学 | A kind of duplicate rows License Plate Segmentation method and system based on CNN |
| CN111046891A (en) * | 2018-10-11 | 2020-04-21 | 杭州海康威视数字技术股份有限公司 | Training method of license plate recognition model, and license plate recognition method and device |
| CN109508715B (en) * | 2018-10-30 | 2022-11-08 | 南昌大学 | License plate positioning and identifying method based on deep learning |
| CN110147709A (en) * | 2018-11-02 | 2019-08-20 | 腾讯科技（深圳）有限公司 | Training method, device, terminal and the storage medium of vehicle attribute model |
| CN109447081B (en) * | 2018-11-14 | 2021-07-23 | 广东小天才科技有限公司 | Method and system for acquiring optical character recognition data set |
| CN109583493A (en) * | 2018-11-27 | 2019-04-05 | 上海交通大学 | A kind of credit card detection and digit recognition method based on deep learning |
| CN109815956B (en) * | 2018-12-28 | 2022-12-09 | 安徽清新互联信息科技有限公司 | License plate character recognition method based on self-adaptive position segmentation |
| CN109858488B (en) * | 2018-12-28 | 2021-09-17 | 众安信息技术服务有限公司 | Handwritten sample recognition method and system based on sample enhancement |
| CN109766805B (en) * | 2018-12-28 | 2022-12-09 | 安徽清新互联信息科技有限公司 | Deep learning-based double-layer license plate character recognition method |
| CN109801375B (en) * | 2019-01-24 | 2021-02-09 | 电子科技大学 | Three-dimensional reconstruction method of porous material based on deep convolution anti-neural network |
| CN109815943B (en) * | 2019-03-18 | 2021-02-09 | 北京石油化工学院 | Hazardous chemical storage stacking picture sample generation method and system |
| CN109993172B (en) * | 2019-03-25 | 2022-12-06 | 西安电子科技大学 | License plate recognition method for road traffic flow |
| CN110245613B (en) * | 2019-06-17 | 2023-01-20 | 珠海华园信息技术有限公司 | Ship board identification method based on deep learning feature comparison |
| CN110309889A (en) * | 2019-07-04 | 2019-10-08 | 西南大学 | A kind of Old-Yi character symbol restorative procedure of double arbiter GAN |
| CN110318731A (en) * | 2019-07-04 | 2019-10-11 | 东北大学 | A kind of oil well fault diagnostic method based on GAN |
| CN110348448A (en) * | 2019-07-05 | 2019-10-18 | 黑龙江科技大学 | A kind of license plate character recognition method based on convolutional neural networks |
| CN110659649A (en) * | 2019-09-29 | 2020-01-07 | 哈尔滨工业大学 | Image processing and character recognition algorithm based on near infrared light imaging |
| CN111046755A (en) * | 2019-11-27 | 2020-04-21 | 上海眼控科技股份有限公司 | Character recognition method, character recognition device, computer equipment and computer-readable storage medium |
| CN110969165B (en) * | 2019-11-28 | 2024-04-09 | 中国科学院半导体研究所 | Handwritten character recognition method, handwritten character recognition device, electronic equipment and storage medium |
| CN111160338B (en) * | 2019-12-24 | 2023-08-01 | 北京奥易克斯科技有限公司 | Fuzzy license plate recognition method |
| CN111310791A (en) * | 2020-01-17 | 2020-06-19 | 电子科技大学 | Dynamic progressive automatic target identification method based on small sample number set |
| CN111401360B (en) * | 2020-03-02 | 2023-06-20 | 杭州雄迈集成电路技术股份有限公司 | Method and system for optimizing license plate detection model, license plate detection method and system |
| CN113553885A (en) * | 2020-04-26 | 2021-10-26 | 复旦大学 | Natural scene text recognition method based on generation countermeasure network |
| CN111639542A (en) * | 2020-05-06 | 2020-09-08 | 中移雄安信息通信科技有限公司 | License plate recognition method, device, equipment and medium |
| CN111612759B (en) * | 2020-05-19 | 2023-11-07 | 佛山科学技术学院 | Printed matter defect identification method based on deep convolution generation type countermeasure network |
| CN111612081B (en) * | 2020-05-25 | 2024-04-02 | 深圳前海微众银行股份有限公司 | Training method, device, equipment and storage medium for recognition model |
| CN111932645B (en) * | 2020-06-12 | 2022-08-26 | 重庆大学 | Method for automatically generating ink and wash painting based on generation countermeasure network GAN |
| CN111881914B (en) * | 2020-06-23 | 2024-02-13 | 安徽清新互联信息科技有限公司 | License plate character segmentation method and system based on self-learning threshold |
| CN113435446B (en) * | 2021-07-07 | 2023-10-31 | 南京云创大数据科技股份有限公司 | Deep learning-based inclined license plate correction method |
| CN114581903A (en) * | 2022-03-24 | 2022-06-03 | 南通大学 | License plate character recognition method based on convolutional neural network |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9785855B2 (en) * | 2015-12-17 | 2017-10-10 | Conduent Business Services, Llc | Coarse-to-fine cascade adaptations for license plate recognition with convolutional neural networks |
-
2017
- 2017-09-02 CN CN201710781905.0A patent/CN107563385B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106951867A (en) * | 2017-03-22 | 2017-07-14 | 成都擎天树科技有限公司 | Face identification method, device, system and equipment based on convolutional neural networks |
Non-Patent Citations (1)
| Title |
|---|
| 基于深度学习神经网络的车牌字符识别技术的研究;王晶;《工业控制计算机》;20170331;第30卷(第3期);51-52 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107563385A (en) | 2018-01-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107563385B (en) | License plate character recognition method based on depth convolution production confrontation network | |
| CN105956560B (en) | A kind of model recognizing method based on the multiple dimensioned depth convolution feature of pondization | |
| CN104463101B (en) | Answer recognition methods and system for character property examination question | |
| CN110610129A (en) | Deep learning face recognition system and method based on self-attention mechanism | |
| CN107766787A (en) | Face character recognition methods, device, terminal and storage medium | |
| CN105138993A (en) | Method and device for building face recognition model | |
| CN106156766A (en) | The generation method and device of line of text grader | |
| CN110210027B (en) | Fine-grained emotion analysis method, device, equipment and medium based on ensemble learning | |
| CN106503616A (en) | A kind of Mental imagery Method of EEG signals classification of the learning machine that transfinited based on layering | |
| CN111898550B (en) | Expression recognition model building method and device, computer equipment and storage medium | |
| CN107506350A (en) | A kind of method and apparatus of identification information | |
| CN108664975A (en) | A kind of hand-written Letter Identification Method of Uighur, system and electronic equipment | |
| CN105912525A (en) | Sentiment classification method for semi-supervised learning based on theme characteristics | |
| CN111401156A (en) | Image identification method based on Gabor convolution neural network | |
| CN109670559A (en) | Recognition methods, device, equipment and the storage medium of handwritten Chinese character | |
| CN105956570A (en) | Lip characteristic and deep learning based smiling face recognition method | |
| Jordao et al. | Deep network compression based on partial least squares | |
| CN114742224A (en) | Pedestrian re-identification method and device, computer equipment and storage medium | |
| CN109255339A (en) | Classification method based on adaptive depth forest body gait energy diagram | |
| CN110414431B (en) | Face recognition method and system based on elastic context relation loss function | |
| CN109145723A (en) | A kind of seal recognition methods, system, terminal installation and storage medium | |
| CN103136546A (en) | Multi-dimension authentication method and authentication device of on-line signature | |
| Guermazi et al. | Facial micro-expression recognition based on accordion spatio-temporal representation and random forests | |
| CN111310820A (en) | Foundation meteorological cloud chart classification method based on cross validation depth CNN feature integration | |
| CN111242114B (en) | Character recognition method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |