CN109977980A - A kind of method for recognizing verification code and device - Google Patents
A kind of method for recognizing verification code and device Download PDFInfo
- Publication number
- CN109977980A CN109977980A CN201711466525.4A CN201711466525A CN109977980A CN 109977980 A CN109977980 A CN 109977980A CN 201711466525 A CN201711466525 A CN 201711466525A CN 109977980 A CN109977980 A CN 109977980A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- identifying code
- training
- networks model
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Character Discrimination (AREA)
Abstract
The embodiment of the present invention provides a kind of method for recognizing verification code and device, for solving the terminal device technical problem lower there are recognition efficiency when identification cracks identifying code.This method comprises: will have markd sample image input convolutional neural networks model, convolutional neural networks model is trained, wherein, include the standardization layer for the characteristic value progress standardization processing to the characteristic information after convolution in the convolutional neural networks model, and the weight quantity of convolution is less than or equal to the weight quantity of convolution in Standard convolution neural network model in the convolutional neural networks model after training, the characteristic value of the characteristic information of the image of extraction is in preset range；By the convolutional neural networks model after identifying code image to be identified input training, the recognition result of identifying code image to be identified is obtained, recognition result is used to indicate at least one the verifying code character in identifying code image included.
Description
Technical field
The present invention relates to field of computer technology, in particular to a kind of method for recognizing verification code and device.
Background technique
Identifying code (CAPTCHA) is " Completely Automated Public Turing test to tell
The abbreviation of Computers and Humans Apart " (the full-automatic turing test for distinguishing computer and the mankind), is a kind of area
Dividing user is computer or the public full auto-programs of people, and the identifying code at present with upper and lower case letter and number is widely used in
Major website, for preventing machine automatic batch from registering and logging in and pour water repeatedly, letter and number itself distorts and with very
More noises and horizontal line.
Identifying code concept is with the fast development of Internet technology and application, network are providing affluent resources and pole to people
While convenience, incident is exactly the safety issue of internet system.The appearance of identifying code, exactly reinforcement web system
The product for safety of uniting.
Identify that the mode cracked includes: for identifying code in the prior art
The first: being based on image procossing mode, and by pretreatment (binaryzation, CFS, connected domain), detection (finds out text institute
Main region), pre-treatment (some rotations are done to image content, distort, segmentation etc.), finally again pass through pattern-recognition or machine
Device learning algorithm is trained, and exports the confidence level of constituent class device to judge which letter or number may be belonged to.
In terms of the segmentation that the technological difficulties of which essentially consist in identifying code picture, for the matching of identification, OCR technique
It is very mature, it is fully available for the identification of identifying code picture, but the complicated most adhesion of identifying code picture, dividing processing
It is more troublesome.
Second: the mode based on deep learning identifies classification verifying by classical picture recognition network model
Content in code picture.But it since classical picture recognition network structure is excessive, needs computing resource more, and needs a large amount of
Sample is marked, convergence rate is very slow, it is difficult to apply in finite time.
It can be seen that there are the lower technologies of recognition efficiency to ask when identification cracks identifying code for terminal device in the prior art
Topic.
Summary of the invention
The embodiment of the present invention provides a kind of method for recognizing verification code and device, for solving terminal device in the prior art
The technical problem lower there are recognition efficiency when identification cracks identifying code.
In a first aspect, the embodiment of the present invention provides a kind of method for recognizing verification code, comprising the following steps:
To have markd sample image input convolutional neural networks model, the convolutional neural networks model will be instructed
Practice；Include the rule for the characteristic value progress standardization processing to the characteristic information after convolution in the convolutional neural networks model
Generalized layer, and the weight quantity of convolution is less than or equal to Standard convolution neural network model in the convolutional neural networks model after training
The characteristic value of the weight quantity of middle convolution, the characteristic information of the image of extraction is in preset range；
By the convolutional neural networks model after identifying code image to be identified input training, obtain described to be identified
The recognition result of identifying code image, the recognition result are used to indicate at least one identifying code in the identifying code image included
Character.
Optionally, will have markd sample image input convolutional neural networks model, to the convolutional neural networks mould
Type is trained, comprising:
To have markd sample image according to training regulation sample size and input the progress of convolutional neural networks model in batches
Training；
It determines that frequency of training reaches preset times, calculates the average loss rate and accuracy of every batch of sample image, it is described
Accuracy is used to characterize the probability that predicted value and mark value match, and the loss late is used to characterize the number of specimen discerning failure
Amount；
It determines that the accuracy is higher than setting value, records the current training parameter of the convolutional neural networks model.
Optionally, determine the accuracy be higher than setting value, record the current training of the convolutional neural networks model
After parameter, before the convolutional neural networks model after identifying code image to be identified input training, the method is also
Include:
By the convolutional neural networks model after test sample collection input training, the test knot of each test sample is determined
Fruit；
Each test result is matched with the mark of corresponding test sample, it is pre- to determine that the matching degree is more than or equal to
If matching degree.
Optionally, the convolutional neural networks model after the input training by identifying code image to be identified, obtains
The recognition result of the identifying code image to be identified, comprising:
By the convolutional neural networks model after the identifying code image input training to be identified, extract described wait know
At least two characteristic informations of other identifying code image；
Standardization processing is carried out to the characteristic value of at least two characteristic information, so that after the standardization processing extremely
The characteristic value of few two characteristic informations is in preset range；
According to the recognition result of treated at least two characteristic informations the determine identifying code image to be identified.
Optionally, according to the identification knot of treated at least two characteristic informations the determine identifying code image to be identified
Fruit, comprising:
The progress of at least two characteristic informations is more to treated for the character quantity in identifying code image extracted as needed
Classification of task obtains the recognition result of the identifying code image.
Second aspect, the embodiment of the present invention provide a kind of verifying code recognition device, comprising:
Character set data module, the character quantity of the identifying code image for input to be arranged, and to the identifying code that will be inputted
Picture is pre-processed, and the pretreatment includes size adjusting and gradation conversion, and pretreated identifying code picture has in advance
If Pixel Dimensions；
Training module, for according to training regulation sample size will there is the sample image of mark to input convolutional Neural in batches
Network model is trained the convolutional neural networks model；Wherein, the convolutional neural networks model after training can
Identify that the identifying code with special characteristic, the convolutional neural networks model include for the feature to the characteristic information after convolution
Value carries out the standardization layer of standardization processing；
Identification module, for obtaining the convolutional neural networks model after identifying code image to be identified input training
The recognition result of the identifying code image to be identified is obtained, the recognition result, which is used to indicate in the identifying code image, includes
At least one verifying code character.
Optionally, the character set data module is also used to:
According to the character quantity in sample image, the character quantity that the convolutional neural networks model needs to identify is set, and
It is pre-processed to by identifying code picture to be identified, the pretreatment includes size adjusting and gradation conversion, and after pretreatment
Identifying code picture have preset Pixel Dimensions.
Optionally, the verifying code recognition device further include:
Test module is connected with the training module, for determining the training module for the convolutional Neural net
When the frequency of training of network model reaches preset times, by have tagged sample image to the convolutional neural networks model into
Row test, and when the accuracy for determining test result is higher than setting value, record the current instruction of the convolutional neural networks model
Practice parameter and model；Wherein, the accuracy is for characterizing the probability that the predicted value of sample image and the value of mark match.
Optionally, the identification module includes:
Characteristic extracting module, for the convolutional neural networks after training the identifying code image input to be identified
Model extracts at least two characteristic informations of the identifying code image to be identified；
Standardize layer, standardization processing is carried out for the characteristic value at least two characteristic information, so that the rule
The characteristic value of generalized treated at least two characteristic informations is in preset range；
Determining module, at least two characteristic informations to determine the identifying code image to be identified according to treated
Recognition result.
Optionally, the determining module is used for:
The progress of at least two characteristic informations is more to treated for the character quantity in identifying code image extracted as needed
Classification of task obtains the recognition result of the identifying code image.
The third aspect, the embodiment of the present invention provide a kind of computer installation, and the computer installation includes processor, described
Method as described in relation to the first aspect is realized when processor is for executing the computer program stored in memory.
Fourth aspect, the embodiment of the present invention provide a kind of computer readable storage medium, the computer-readable storage medium
Matter is stored with computer instruction, when described instruction is run on computers, so that computer executes as described in relation to the first aspect
Method.
In the embodiment of the present invention, convolutional neural networks model is inputted by that will have markd sample image, to convolution mind
It is trained through network model, the convolutional neural networks model after training is enabled to identify the identifying code with special characteristic,
Since the convolutional neural networks model includes standardization layer, and the weight quantity of convolution is less than or equal in convolutional neural networks model
The weight quantity of convolution in the conventional model of standard, and standardized by the layer that standardizes to the characteristic information after convolution,
So that the characteristic value of the characteristic information of model extraction is in preset range, the quantity of convolutional layer and the number of weight are effectively reduced
Amount improves recognition accuracy to greatly speed up the convergence time of model training.
Detailed description of the invention
Fig. 1 is the architecture diagram of the convolutional neural networks model provided in the embodiment of the present invention；
Fig. 2 is the flow chart of method for recognizing verification code in the embodiment of the present invention；
Fig. 3 is the structure chart that code recognition device is verified in the embodiment of the present invention；
Fig. 4 is the structure chart of computer installation in the embodiment of the present invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention make into
It is described in detail to one step, it is clear that the described embodiments are only some of the embodiments of the present invention, rather than whole implementation
Example.Based on the embodiments of the present invention, obtained by those of ordinary skill in the art without making creative efforts
All other embodiment, shall fall within the protection scope of the present invention.
The verifying code recognition device that technical solution described herein can be used for having identifying code identification function, the identifying code
Identification device can be set in other electronic equipments such as terminal or server.
Firstly, introducing convolutional neural networks model employed in the embodiment of the present invention in conjunction with attached drawing.
Fig. 1 is the configuration diagram of convolutional neural networks model in the embodiment of the present invention.The convolutional neural networks model phase
For existing convolutional neural networks model, standardization layer is increased between convolutional layer, which is used for volume
The characteristic value of characteristic information after product carries out standardization processing.Meanwhile the convolutional neural networks model in the embodiment of the present invention is also
Multitask classifier is increased in network output layer, is shown together in Fig. 1, i.e. multitask classifier layer.
Specifically, the convolutional neural networks model in the embodiment of the present invention can reduce parameter by the way that standardization layer is added
Amount and the quantity of convolutional layer accelerate the convergence speed of network model to reduce the size and network depth of convolution kernel.Together
When, in the output layer of network model, multitask classifier layer is carried out according to the character quantity in the picture of desired extraction, is increased
The versatility of identifying code identification.
In the embodiment of the present invention, before stating convolutional neural networks model progress identifying code image recognition in use, need to lead to
A certain number of sample images are crossed to be trained it.
When being trained preparation, the character quantity in the identifying code image identified can will be needed to be defined first.Example
Such as the conversion of vector coding can be carried out to the character of identical quantity, when to train more by customized steering volume mode
Good carry out feature extraction, so that the similarity between character improves, it is made of 2 two-way modules, i.e. character changes into vector,
Character is changed by vector again.
Wherein, steering volume processing, which refers to, is converted into numerical value vector mode, such as one-hot coding (solely heat for each character
Programming) or other coding modes, in order to embody the similarity between character.For example, if phase between letter A and letter b
It adjusts the distance bigger than the relative distance between alphabetical A and letter C, then can be embodied from the vector value after coding.
In turn, it needs to be adjusted the identifying code image size of input and gradation conversion, by the identifying code of input model
Image Adjusting is to defined Pixel Dimensions.Since convolutional neural networks model is when carrying out character recognition, picture is not considered generally
Color information therefore also the rgb color image of multichannel can be changed into grayscale image.
Then, will have tagged sample image to divide, and can specifically be divided into training sample set, verifying sample set
And test sample collection totally 3 parts, and training sample set is provided, verify the size of the batch testing collection of sample set.Wherein,
Supervised learning training of the training sample set for model, by backpropagation, adjusting parameter learns；Verifying sample set can be used
In model in learning process, carrying out interim detection is optimal model performance for finely tuning the hyper parameter of model；It surveys
Examination sample set can be used for examining it based on the accuracy rate of more typically data after model reaches certain accuracy rate.
The allocation proportion of training sample set, verifying sample set and test sample collection can be 8:1:1.That is 3 kinds
Sample set is that the tagged sample pattern of tool can upset all sample images, division result may is that when dividing at random
Training sample set accounts for 80%, verifies sample set and test sample collection respectively accounts for 10%.
Then, equipment can read each sample set and input model, so that model identifies the sample of wherein marking error
This image, and respective handling is done, such as discard processing.
In addition it is also necessary to which the framework to convolutional neural networks model is arranged accordingly, weight and offset parameter are set
Initial value.The convolutional neural networks model of setting can be such that the treatment process of identifying code image
1) convolution kernel for using larger size, extracts for large range of feature in identifying code picture；
2) the maximization pond layer for choosing suitable dimension carries out feature and simplifies；
3) the volume collection core for using smaller size, expands the quantity of convolution kernel；
4) it carries out maximizing Chi Hualai progress characteristic details feature extraction and simplify, to extract more small features letters
Breath, and the number of single channel picture neuron is reduced, thus the weight and offset of training needed for reducing；
5) standardization processing is criticized, by neuron weight and offset standardization in certain distribution, is effectively reduced anti-
To propagate when, it is understood that there may be gradient disperse.
6) it is carried out continuously 2 layers of convolution, while continuing to expand the quantity of convolution kernel, further decreases single channel picture nerve
The number of member, and extract the feature of more details；
7) an and then full articulamentum for being fully deployed mode, access is by certain probability dropping neuron after this
Anti- over-fitting unit；
8) a certain number of character classifiers are accessed, this classifier can be adjusted output according to the variation of character quantity,
It can define the classification of multiple tasks.
Due to joined standardization, initial learning rate can be set bigger, which refers to beginning ladder
The step-length for spending decline can be in advance according to experiment experience setting.In turn, after starting propagated forward and back-propagating, according to
Loss late, as model propagated forward it is calculated value mark target value between difference, such as Euclidean distance or intersect entropy,
Slowly decayed.Loss function uses the mean value costing bio disturbance mode of batch sample.When training, trained stated number is randomly selected
The sample of amount carries out propagated forward calculating, and average loss late and accuracy are calculated by every batch of.Carrying out certain number
Training step after, will do it one-time authentication step such as 100 times, when verification step, and calculate loss late and accuracy, when
When accuracy is higher than setting value, the training parameter and model of entire model are kept, and exit.Test module, training before obtaining
Good model parameter and framework, is predicted using ready test sample collection, and by test result and the result of mark into
Row compares, and obtains the final test accuracy of sample batch verifying.During actual test, model is every time only to single sample
This image is operated.
In practical applications, it in the framework of convolutional neural networks model as shown in Figure 1, sees from left to right, convolution mind
After network model alternately extracts the characteristic information of the image of input by convolutional layer and pond layer, and then pass through batch rule
Generalized module carries out batch standardization processing to the characteristic value of characteristic information, so that the characteristic value of characteristic information is in preset range
Interior, which can be [0.5,1].Neuron weight and offset standardization can be centainly distributed by standardization processing
In range, thus when effectively reducing backpropagation, it is understood that there may be gradient disperse.Such as the characteristic value of the characteristic information of Chi Huahou
Range may be [0.5,3.5], carry out standardization processing after, range locating for characteristic value can be [0.5,1].
As shown in Fig. 2, the embodiment of the present invention also provides a kind of method for recognizing verification code, it is applied to verifying code recognition device,
It include convolutional neural networks model in the verifying code recognition device, which has knot as shown in Figure 1
Structure.This method can be described as follows:
S11: will have markd sample image input convolutional neural networks model, convolutional neural networks model will be instructed
Practice；Wherein, comprising carrying out standardization processing for the characteristic value to the characteristic information after convolution in convolutional neural networks model
Standardize layer, and the weight quantity of convolution is less than or equal to Standard convolution neural network mould in the convolutional neural networks model after training
The characteristic value of the weight quantity of convolution in type, the characteristic information of the image of extraction is in preset range.
Specifically, Standard convolution neural network model can refer to existing conventional convolutional neural networks model, i.e.,
It is not provided with the network model of standardization layer and multitask classifier layer.Sample image can be the verifying of pre-prepd mark
Code image, the mark of sample image can indicate the relevant information of identifying code in sample image, such as specific verifying code word
Symbol and/or character quantity etc..Standardization layer in convolutional neural networks model can be used for the feature diagram data after convolution into
Row standardization processing is referred to as " more herein in addition, convolutional neural networks model can also include classifier in output layer
Classification of task device layer " can be adjusted according to the variation of character quantity in classifier layer and be exported.
In the embodiment of the present invention, verifying code recognition device can be trained convolutional neural networks model based on sample image
When, process can be such that
Firstly, verifying code recognition device pre-processes sample image, which includes that size adjusting and gray scale turn
It changes, pretreated identifying code picture has preset Pixel Dimensions.
In turn, verifying code recognition device will have markd sample image according to sample size as defined in training and input in batches
Convolutional neural networks model is trained, such as every batch of includes 80 sample images.Its training process refers to aforementioned to volume
The framework of product neural network model introduces content, and details are not described herein again.
In turn, if it is determined that frequency of training reaches preset times, then verifies code recognition device and calculate every batch of sample image
Average loss rate and accuracy, the probability which can be used for characterizing predicted value and mark value matches, the loss late
For characterizing the quantity of specimen discerning failure.
If it is determined that accuracy is higher than setting value, then verifying code recognition device, to can record convolutional neural networks model current
Training parameter and model can exit training at this time.
In practical applications, after being trained to convolutional neural networks model, before being put into use can also to its into
Row test.At this point, test sample collection can be inputted to the convolutional neural networks model after training, the test of each test sample is determined
As a result, each test result is matched with the mark of corresponding test sample in turn.
If it is determined that matching degree is more than or equal to preset matching degree, such as 80%, the then convolutional neural networks after showing training
Model accuracy with higher can be used for subsequent identifying code image recognition.Certainly, if matching degree is lower than preset matching
Degree, then can continue through sample image and be trained to model, to obtain more preferred training parameter.
Since the convolutional neural networks model in the embodiment of the present invention includes standardization and multitask classification, convolutional Neural net
Network model can reduce the quantity of parameter amount and convolutional layer by the way that standardization is added, to reduce the size and network depth of convolution kernel
Degree, accelerates the convergence speed of network model.It, can be according to the picture of desired extraction meanwhile in the output layer of network model
In character quantity carry out multitask classification, increase the versatility of identifying code identification, therefore reduce training convolutional nerve net
Under the quantity of required sample when network model, such as convolutional neural networks model structure in embodiments of the present invention, it is only necessary to
80% or more recognition accuracy can be realized in the sample patterns of 10000 marks, to the accuracy rate of identifying code image recognition compared with
It is high.
S12: by the convolutional neural networks model after identifying code image to be identified input training, verifying to be identified is obtained
The recognition result of code image, recognition result are used to indicate at least one the verifying code character in identifying code image included.
After determining training parameter and model, verifying code recognition device can pass through the convolutional neural networks model after training
Identifying code image is identified.The convolutional neural networks model after training identifying code image to be identified input
When, verifying code recognition device can first pre-process identifying code image to be identified, by the identifying code picture of input model
Defined Pixel Dimensions are adjusted to, while also identifying code image to be identified can be converted to grayscale image.
And then the characteristic information of identifying code image to be identified is extracted by convolutional layer and pond layer.For example, in conjunction in Fig. 1
Convolutional neural networks model, first can be used larger size convolution kernel, for large range of feature in identifying code picture
It extracts, the maximization pond layer further progress feature for choosing suitable dimension later is simplified.And then reduce volume collection core
Size expands the quantity of volume collection core, and further progress maximizes Chi Hualai and carries out characteristic details feature extraction and simplify, to extract
More tiny characteristics information out, and reduce the number of single channel picture neuron, thus the weight of training needed for reducing and partially
It moves, while improving the precision of characteristic information.
Then, verifying code recognition device carries out standardization processing to the characteristic value of at least two characteristic informations, so that specification
The characteristic value for changing treated at least two characteristic informations be in preset range, such as the feature of the characteristic information after standardizing
Value is in the range of [0.5,1].By standardization, neuron weight and offset can be standardized in certain distribution,
Effectively reduce gradient disperse that may be present when backpropagation.
Further, verifying code recognition device can be by being carried out continuously 2 layers of convolutional layer, while continuing to expand convolution kernel
Quantity further decreases the number of single channel picture neuron, and extracts the feature of more details.And then, completely by one
The full articulamentum of expansion mode also can access the anti-over-fitting unit by certain probability dropping neuron after full articulamentum.
A certain number of character classifiers are finally accessed, this classifier can be adjusted output, Ji Keding according to the variation of character quantity
The classification of adopted multiple tasks, so that the recognition result of identifying code image to be identified, such as the identifying code identified are obtained, such as
4 alphabetic characters etc..
In the embodiment of the present invention, convolutional neural networks model is inputted by that will have markd sample image, to convolution mind
It is trained through network model, the convolutional neural networks model after training is enabled to identify the identifying code with special characteristic,
Since the convolutional neural networks model includes standardization layer and classifier layer, and in convolutional neural networks model convolution weight number
Amount is less than or equal to the weight quantity of the conventional convolutional neural networks model for not including standardization layer, and the characteristic value of extraction is in pre-
If in range, model is allowed to identify more multicharacter identifying code, simultaneously because standardization layer is to the characteristic pattern number after convolution
According to being standardized, the quantity of convolutional layer and the quantity of weight are effectively reduced, thus when greatly speeding up the convergence of model training
Between, improve recognition accuracy.
As described in Figure 3, the embodiment of the present invention also provides a kind of verifying code recognition device, which can be used to implement Fig. 2
Shown in method, the apparatus may include character set data module 21, training module 22 and identification modules 23.
Character set data module 21 can be used for being arranged the character quantity in the identifying code image of input, and to will input
Identifying code picture is pre-processed, and the pretreatment includes size adjusting and gradation conversion, and pretreated identifying code picture
With preset Pixel Dimensions.
Training module 22 can be used for according to training regulation sample size there is the sample image of mark to input volume in batches
Product neural network model, is trained the convolutional neural networks model；Wherein, the convolutional neural networks mould after training
Type can identify that the identifying code with special characteristic, the convolutional neural networks model include for the characteristic information after convolution
Characteristic value carry out standardization processing standardization layer.
Identification module 23 can be used for inputting identifying code image to be identified into the convolutional neural networks mould after training
Type, obtains the recognition result of the identifying code image to be identified, and the recognition result is used to indicate in the identifying code image
Including at least one verifying code character.
Optionally, the character set data module 21 is also used to: the volume is arranged according to the character quantity in sample image
The character quantity that product neural network model needs to identify, and pre-processed to by identifying code picture to be identified, the pre- place
Reason includes size adjusting and gradation conversion, and pretreated identifying code picture has preset Pixel Dimensions.
Optionally, the verifying code recognition device further include: the test module being connected with the training module, for true
It is tagged by having when the fixed training module reaches preset times for the frequency of training of the convolutional neural networks model
Sample image tests the convolutional neural networks model, and when the accuracy for determining test result is higher than setting value,
Record the convolutional neural networks model current training parameter and model；Wherein, the accuracy is for characterizing sample image
Predicted value and mark the probability that matches of value.
Optionally, the identification module 23 includes:
Characteristic extracting module, for the convolutional neural networks after training the identifying code image input to be identified
Model extracts at least two characteristic informations of the identifying code image to be identified；
Normalizing block carries out standardization processing for the characteristic value at least two characteristic information, so that described
The characteristic value of at least two characteristic informations after standardization processing is in preset range；
Determining module, at least two characteristic informations to determine the identifying code image to be identified according to treated
Recognition result.
Optionally, the determining module is used for:
The progress of at least two characteristic informations is more to treated for the character quantity in identifying code image extracted as needed
Classification of task obtains the recognition result of the identifying code image.
As shown in figure 4, also providing a kind of computer installation computer installation in the embodiment of the present invention includes processor 31
With memory 32, wherein processor 31 is for realizing the embodiment of the present invention when executing the computer program stored in memory 32
The step of method of the processing updating digital certificate request provided in one.
Optionally, processor 31 specifically can be central processing unit, application-specific integrated circuit (Application
Specific Integrated Circuit, ASIC), it can be one or more integrated circuits executed for controlling program,
It can be the hardware circuit of use site programmable gate array (Field Programmable Gate Array, FPGA) exploitation,
It can be baseband processor.
Optionally, processor 31 may include at least one processing core.
Optionally, electronic equipment further includes memory 32, and memory 32 may include read-only memory (Read Only
Memory, ROM), random access memory (Random Access Memory, RAM) and magnetic disk storage.Memory 32 is used for
The data required when running of storage processor 31.The quantity of memory 32 is one or more.
In an alternative embodiment of the invention, a kind of computer readable storage medium is also provided, the computer-readable storage medium
Matter is stored with computer instruction, is may be implemented when computer instruction is run on computers as the present invention implements an example offer
The step of handling the method for updating digital certificate request.
In embodiments of the present invention, it should be understood that the method and server of disclosed processing updating digital certificate request,
It may be implemented in other ways.For example, apparatus embodiments described above are merely indicative, for example, unit
It divides, only a kind of logical function partition, there may be another division manner in actual implementation, such as multiple units or components
It can be combined or can be integrated into another system, or some features can be ignored or not executed.Another point, it is shown or
The mutual coupling, direct-coupling or communication connection discussed can be through some interfaces, the indirect coupling of equipment or unit
It closes or communicates to connect, can be electrical or other forms.
Each functional unit in embodiments of the present invention can integrate in one processing unit or each unit can also
To be independent physical module.
It, can if integrated unit is realized in the form of SFU software functional unit and when sold or used as an independent product
To be stored in a computer readable storage medium.Based on this understanding, the technical solution of the embodiment of the present invention is complete
Portion or part can be embodied in the form of software products, which is stored in a storage medium, packet
It includes some instructions to use so that a computer equipment, such as can be personal computer, server or the network equipment etc.,
Or processor (Processor) executes all or part of the steps of the method for each embodiment of the present invention.And storage above-mentioned is situated between
Matter includes: general serial bus USB (Universal Serial Bus flash drive, USB), mobile hard disk, read-only
Memory (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disk or
The various media that can store program code such as person's CD.
Above embodiments are only used for that technical solution of the present invention is described in detail, but the explanation of above embodiments is only
It is the method for being used to help understand the embodiment of the present invention, should not be construed as the limitation to the embodiment of the present invention.The art
Any changes or substitutions that can be easily thought of by technical staff, should all cover within the protection scope of the embodiment of the present invention.
Claims (12)
1. a kind of method for recognizing verification code is applied to verifying code recognition device characterized by comprising
To have markd sample image input convolutional neural networks model, the convolutional neural networks model will be trained；
Wherein, comprising carrying out standardization processing for the characteristic value to the characteristic information after convolution in the convolutional neural networks model
Standardize layer, and the weight quantity of convolution is less than or equal to Standard convolution neural network mould in the convolutional neural networks model after training
The characteristic value of the weight quantity of convolution in type, the characteristic information of the image of extraction is in preset range；
By the convolutional neural networks model after identifying code image to be identified input training, the verifying to be identified is obtained
The recognition result of code image, the recognition result are used to indicate at least one the verifying code word in the identifying code image included
Symbol.
2. the method as described in claim 1, which is characterized in that will have markd sample image input convolutional neural networks mould
Type is trained the convolutional neural networks model, comprising:
To have markd sample image according to training regulation sample size and input convolutional neural networks model in batches and is trained；
It determines that frequency of training reaches preset times, calculates the average loss rate and accuracy of every batch of sample image, it is described correct
Rate is used to characterize the probability that predicted value and mark value match, and the loss late is used to characterize the quantity of specimen discerning failure；
It determines that the accuracy is higher than setting value, records the current training parameter of the convolutional neural networks model.
3. method according to claim 2, which is characterized in that determine the accuracy be higher than setting value, record the volume
After the current training parameter of product neural network model, by the convolutional Neural after identifying code image to be identified input training
Before network model, the method also includes:
By the convolutional neural networks model after test sample collection input training, the test result of each test sample is determined；
Each test result is matched with the mark of corresponding test sample, determines that the matching degree is more than or equal to default
With degree.
4. the method as described in claim 1, which is characterized in that the institute after the input training by identifying code image to be identified
Convolutional neural networks model is stated, the recognition result of the identifying code image to be identified is obtained, comprising:
By the convolutional neural networks model after the identifying code image input training to be identified, extract described to be identified
At least two characteristic informations of identifying code image；
Standardization processing is carried out to the characteristic value of at least two characteristic information, so that at least two after the standardization processing
The characteristic value of a characteristic information is in preset range；
According to the recognition result of treated at least two characteristic informations the determine identifying code image to be identified.
5. method as claimed in claim 4, which is characterized in that according to treated at least two characteristic informations determine it is described to
The recognition result of the identifying code image of identification, comprising:
To treated, at least two characteristic informations carry out multitask to the character quantity in identifying code image extracted as needed
Classification, obtains the recognition result of the identifying code image.
6. a kind of verifying code recognition device characterized by comprising
Character set data module, the character quantity of the identifying code image for input to be arranged, and to the identifying code picture that will be inputted
It is pre-processed, the pretreatment includes size adjusting and gradation conversion, and pretreated identifying code picture is with preset
Pixel Dimensions；
Training module, for according to training regulation sample size will there is the sample image of mark to input convolutional neural networks in batches
Model is trained the convolutional neural networks model；Wherein, the convolutional neural networks model after training can identify
Identifying code with special characteristic, the convolutional neural networks model include for the characteristic value to the characteristic information after convolution into
The standardization layer of row standardization processing；
Identification module, for obtaining institute for the convolutional neural networks model after identifying code image to be identified input training
The recognition result of identifying code image to be identified is stated, it includes at least that the recognition result, which is used to indicate in the identifying code image,
One verifying code character.
7. device as claimed in claim 6, which is characterized in that the character set data module is also used to:
The character quantity that the convolutional neural networks model needs to identify is set according to the character quantity in sample image, and to general
Identifying code picture to be identified is pre-processed, and the pretreatment includes size adjusting and gradation conversion, and pretreated is tested
Demonstrate,proving code picture has preset Pixel Dimensions.
8. device as claimed in claim 7, which is characterized in that the verifying code recognition device further include:
Test module is connected with the training module, for determining the training module for the convolutional neural networks mould
When the frequency of training of type reaches preset times, the convolutional neural networks model is surveyed by having tagged sample image
Examination, and when the accuracy for determining test result is higher than setting value, record the current training ginseng of the convolutional neural networks model
Several and model；Wherein, the accuracy is for characterizing the probability that the predicted value of sample image and the value of mark match.
9. device as claimed in claim 8, which is characterized in that the identification module includes:
Characteristic extracting module, for the convolutional neural networks mould after training the identifying code image input to be identified
Type extracts at least two characteristic informations of the identifying code image to be identified；
Standardize layer, standardization processing is carried out for the characteristic value at least two characteristic information, so that the standardization
The characteristic value of treated at least two characteristic informations is in preset range；
Determining module, for the identification according to treated at least two characteristic informations the determine identifying code image to be identified
As a result.
10. device as claimed in claim 9, which is characterized in that the determining module is used for:
To treated, at least two characteristic informations carry out multitask to the character quantity in identifying code image extracted as needed
Classification, obtains the recognition result of the identifying code image.
11. a kind of computer installation, which is characterized in that the computer installation includes processor, and the processor is for executing
It is realized when the computer program stored in memory such as any claim the method in claim 1-5.
12. a kind of computer readable storage medium, which is characterized in that the computer-readable recording medium storage has computer to refer to
It enables, when described instruction is run on computers, so that computer executes the side as described in claim any in claim 1-5
Method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711466525.4A CN109977980A (en) | 2017-12-28 | 2017-12-28 | A kind of method for recognizing verification code and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711466525.4A CN109977980A (en) | 2017-12-28 | 2017-12-28 | A kind of method for recognizing verification code and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109977980A true CN109977980A (en) | 2019-07-05 |
Family
ID=67075297
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711466525.4A Pending CN109977980A (en) | 2017-12-28 | 2017-12-28 | A kind of method for recognizing verification code and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109977980A (en) |
Cited By (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110555462A (en) * | 2019-08-02 | 2019-12-10 | 深圳索信达数据技术有限公司 | non-fixed multi-character verification code identification method based on convolutional neural network |
| CN110956177A (en) * | 2019-11-22 | 2020-04-03 | 成都市映潮科技股份有限公司 | Hybrid verification code identification method and system |
| CN111027568A (en) * | 2019-10-30 | 2020-04-17 | 中国科学院自动化研究所南京人工智能芯片创新研究院 | Iterative neural network batch normalization system |
| CN111143213A (en) * | 2019-12-24 | 2020-05-12 | 北京数衍科技有限公司 | Software automation test method and device and electronic equipment |
| CN111259366A (en) * | 2020-01-22 | 2020-06-09 | 支付宝(杭州)信息技术有限公司 | Verification code recognizer training method and device based on self-supervision learning |
| CN111275039A (en) * | 2020-01-17 | 2020-06-12 | 深圳信息职业技术学院 | Water gauge character positioning method and device, computing equipment and storage medium |
| CN111368824A (en) * | 2020-02-24 | 2020-07-03 | 河海大学常州校区 | Instrument identification method, mobile device and storage medium |
| CN111611988A (en) * | 2020-05-22 | 2020-09-01 | 上海携程商务有限公司 | Picture verification code identification method and device, electronic equipment and computer readable medium |
| CN111652220A (en) * | 2020-06-04 | 2020-09-11 | 上海鸢安智能科技有限公司 | Metal part identity recognition method, system, storage medium and terminal based on dot matrix image |
| CN111783064A (en) * | 2020-06-30 | 2020-10-16 | 平安国际智慧城市科技股份有限公司 | Method and device for generating graphic verification code, computer equipment and storage medium |
| CN112016591A (en) * | 2020-08-04 | 2020-12-01 | 杰创智能科技股份有限公司 | Training method of image recognition model and image recognition method |
| CN112989312A (en) * | 2020-11-30 | 2021-06-18 | 北京金堤科技有限公司 | Verification code identification method and device, electronic equipment and storage medium |
| WO2022156552A1 (en) * | 2021-01-19 | 2022-07-28 | 北京嘀嘀无限科技发展有限公司 | Method for encrypting verification code image, and device, storage medium and computer program product |
| CN117132989A (en) * | 2023-10-23 | 2023-11-28 | 山东大学 | Character verification code identification method, system and equipment based on convolutional neural network |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184226A (en) * | 2015-08-11 | 2015-12-23 | 北京新晨阳光科技有限公司 | Digital identification method, digital identification device, neural network training method and neural network training device |
| CN106570477A (en) * | 2016-10-28 | 2017-04-19 | 中国科学院自动化研究所 | Vehicle model recognition model construction method based on depth learning and vehicle model recognition method based on depth learning |
| CN107153817A (en) * | 2017-04-29 | 2017-09-12 | 深圳市深网视界科技有限公司 | Pedestrian's weight identification data mask method and device |
-
2017
- 2017-12-28 CN CN201711466525.4A patent/CN109977980A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105184226A (en) * | 2015-08-11 | 2015-12-23 | 北京新晨阳光科技有限公司 | Digital identification method, digital identification device, neural network training method and neural network training device |
| CN106570477A (en) * | 2016-10-28 | 2017-04-19 | 中国科学院自动化研究所 | Vehicle model recognition model construction method based on depth learning and vehicle model recognition method based on depth learning |
| CN107153817A (en) * | 2017-04-29 | 2017-09-12 | 深圳市深网视界科技有限公司 | Pedestrian's weight identification data mask method and device |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110555462A (en) * | 2019-08-02 | 2019-12-10 | 深圳索信达数据技术有限公司 | non-fixed multi-character verification code identification method based on convolutional neural network |
| CN111027568A (en) * | 2019-10-30 | 2020-04-17 | 中国科学院自动化研究所南京人工智能芯片创新研究院 | Iterative neural network batch normalization system |
| CN110956177A (en) * | 2019-11-22 | 2020-04-03 | 成都市映潮科技股份有限公司 | Hybrid verification code identification method and system |
| CN111143213A (en) * | 2019-12-24 | 2020-05-12 | 北京数衍科技有限公司 | Software automation test method and device and electronic equipment |
| CN111275039A (en) * | 2020-01-17 | 2020-06-12 | 深圳信息职业技术学院 | Water gauge character positioning method and device, computing equipment and storage medium |
| CN111275039B (en) * | 2020-01-17 | 2023-05-16 | 深圳信息职业技术学院 | Water gauge character positioning method, device, computing equipment and storage medium |
| CN111259366A (en) * | 2020-01-22 | 2020-06-09 | 支付宝(杭州)信息技术有限公司 | Verification code recognizer training method and device based on self-supervision learning |
| CN111368824B (en) * | 2020-02-24 | 2022-09-23 | 河海大学常州校区 | Instrument identification method, mobile device and storage medium |
| CN111368824A (en) * | 2020-02-24 | 2020-07-03 | 河海大学常州校区 | Instrument identification method, mobile device and storage medium |
| CN111611988A (en) * | 2020-05-22 | 2020-09-01 | 上海携程商务有限公司 | Picture verification code identification method and device, electronic equipment and computer readable medium |
| CN111652220A (en) * | 2020-06-04 | 2020-09-11 | 上海鸢安智能科技有限公司 | Metal part identity recognition method, system, storage medium and terminal based on dot matrix image |
| CN111783064A (en) * | 2020-06-30 | 2020-10-16 | 平安国际智慧城市科技股份有限公司 | Method and device for generating graphic verification code, computer equipment and storage medium |
| CN112016591A (en) * | 2020-08-04 | 2020-12-01 | 杰创智能科技股份有限公司 | Training method of image recognition model and image recognition method |
| CN112989312A (en) * | 2020-11-30 | 2021-06-18 | 北京金堤科技有限公司 | Verification code identification method and device, electronic equipment and storage medium |
| CN112989312B (en) * | 2020-11-30 | 2024-04-30 | 北京金堤科技有限公司 | Verification code identification method and device, electronic equipment and storage medium |
| WO2022156552A1 (en) * | 2021-01-19 | 2022-07-28 | 北京嘀嘀无限科技发展有限公司 | Method for encrypting verification code image, and device, storage medium and computer program product |
| CN117132989A (en) * | 2023-10-23 | 2023-11-28 | 山东大学 | Character verification code identification method, system and equipment based on convolutional neural network |
| CN117132989B (en) * | 2023-10-23 | 2024-01-26 | 山东大学 | Character verification code identification method, system and equipment based on convolutional neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109977980A (en) | A kind of method for recognizing verification code and device | |
| CN109829430B (en) | Cross-modal pedestrian re-identification method and system based on heterogeneous hierarchical attention mechanism | |
| CN105303179A (en) | Fingerprint identification method and fingerprint identification device | |
| CN110717554B (en) | Image recognition method, electronic device, and storage medium | |
| CN106951825A (en) | A kind of quality of human face image assessment system and implementation method | |
| CN108985232A (en) | Facial image comparison method, device, computer equipment and storage medium | |
| EP3588381A1 (en) | Method and apparatus for training classification model, method and apparatus for classifying | |
| CN106663203B (en) | Living body finger print recognition methods and device | |
| CN108922560A (en) | A kind of city noise recognition methods based on interacting depth neural network model | |
| CN108776774A (en) | A kind of human facial expression recognition method based on complexity categorization of perception algorithm | |
| US11062120B2 (en) | High speed reference point independent database filtering for fingerprint identification | |
| CN109766277A (en) | A kind of software fault diagnosis method based on transfer learning and DNN | |
| CN109165639B (en) | Finger vein identification method, device and equipment | |
| CN111783505A (en) | Method and device for identifying forged faces and computer-readable storage medium | |
| CN110909807A (en) | Network verification code identification method and device based on deep learning and computer equipment | |
| CN106295591A (en) | Gender identification method based on facial image and device | |
| CN109271915B (en) | Anti-counterfeiting detection method and device, electronic equipment and storage medium | |
| CN107358209A (en) | Training method, device and method for detecting human face, the device of Face datection model | |
| CN109934255A (en) | A kind of Model Fusion method for delivering object Classification and Identification suitable for beverage bottle recycling machine | |
| CN105117708A (en) | Facial expression recognition method and apparatus | |
| CN107798351A (en) | A kind of personal identification method and system based on deep learning neutral net | |
| CN110134961A (en) | Processing method, device and the storage medium of text | |
| CN106897700B (en) | Single-sample face recognition method and system | |
| CN109189965A (en) | Pictograph search method and system | |
| CN108932533A (en) | Identification model construction method and device, character identifying 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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190705 |