CN107360137A - Construction method and device for the neural network model of identifying code identification - Google Patents
Construction method and device for the neural network model of identifying code identification Download PDFInfo
- Publication number
- CN107360137A CN107360137A CN201710453765.4A CN201710453765A CN107360137A CN 107360137 A CN107360137 A CN 107360137A CN 201710453765 A CN201710453765 A CN 201710453765A CN 107360137 A CN107360137 A CN 107360137A
- Authority
- CN
- China
- Prior art keywords
- identifying code
- neural network
- character
- network model
- identification
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/14—Network architectures or network communication protocols for network security for detecting or protecting against malicious traffic
- H04L63/1441—Countermeasures against malicious traffic
-
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L63/00—Network architectures or network communication protocols for network security
- H04L63/08—Network architectures or network communication protocols for network security for authentication of entities
Landscapes
- Engineering & Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Computer Hardware Design (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Image Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Character Discrimination (AREA)
- Biomedical Technology (AREA)
Abstract
The present invention relates to a kind of construction method, device, storage medium and the computer equipment of the neural network model for identifying code identification.The above method includes：Obtain multiple training identifying codes；Character feature data in each training identifying code of extraction；Handled training the character feature data in identifying code to import in the neural network model comprising initial neural network parameter, generation identification character；The identification error rate of neural network model is calculated with the checking character of corresponding training identifying code according to each identification character；According to identification character, training identifying code and identification error rate adjusting and optimizing neural network model, until identification error rate is less than error-rate threshold；The neural network parameter of neural network model of the neural network parameter as constructed by after last time is adjusted.Construction method, device, storage medium and the computer equipment of the above-mentioned neural network model for identifying code identification, the efficiency of identifying code identification can be effectively improved.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind of neural network model for identifying code identification
Construction method, device, storage medium and computer equipment and a kind of method for recognizing verification code, device, storage medium and computer
Equipment.
Background technology
Identifying code is the behaviors such as the automatic registration in order to prevent malice, digital or alphabetical is given birth to by what is randomly generated for a string
Into picture, and some disturbing factors would generally be added in picture, such as interfering line and Characters Stuck.User's naked eyes identification is tested
After demonstrate,proving code, it need to correctly fill in identifying code and submit website authentication, corresponding operation could be completed after being proved to be successful.Climbed in network
Worm is collected in the running of data, is frequently encountered the situation for needing input validation code.Web crawlers is otherwise known as webpage spider
Spider or network robot etc., it is a kind of program or script that web message is automatically captured according to certain rule.Due to
The artificial checking code efficiency that is manually entered is low, therefore automatic identification identifying code is particularly important.
However, easily the interfering line in by identifying code and Characters Stuck disturb traditional method for recognizing verification code, and
There is limitation in the character species that can be identified, so as to cause identifying code discrimination relatively low.
The content of the invention
The embodiment of the present invention provides a kind of construction method of the neural network model for identifying code identification, device, storage
Medium and computer equipment, can effectively solve the relatively low technical problem of identifying code recognition efficiency, make identifying code identification more accurate
Really.
The embodiment of the present invention also provides a kind of method for recognizing verification code, device, storage medium and computer equipment, Ke Yiyou
Effect solves the relatively low technical problem of identifying code recognition efficiency, makes identifying code identification more accurate.
A kind of construction method of neural network model for identifying code identification, methods described include：
Multiple training identifying codes are obtained, each train in identifying code includes a character to be identified；
Character feature data in each training identifying code of extraction；
Character feature data in the training identifying code are imported to the neutral net for including initial neural network parameter
Handled in model, generation identification character；
The identification of the neural network model is calculated with the checking character of corresponding training identifying code according to each identification character
Error rate；
When the identification error rate is more than error-rate threshold, according to the identification character and the training identifying code adjustment
The neural network parameter of the neural network model, using the neural network parameter after adjustment as initial neural network parameter,
And return to the neutral net mould that the character feature data in the training identifying code are imported and include initial neural network parameter
The step of being handled in type, generating identification character continues to train, until the identification error rate is less than error-rate threshold；Will most
Neural network parameter of the neural network parameter as constructed neural network model after once adjusting afterwards.
In one of the embodiments, the character feature data include every row target pixel points and and fixed row mesh
The sum of pixel is marked, the target pixel points are the pixels that the character to be identified occupies in corresponding training identifying code
Point；
Character feature data in the training identifying code are imported to the neutral net for including initial neural network parameter
Handled in model, generation identification character, including：
By the often row target pixel points and and fixed row target pixel points and import and include initial nerve net
Handled in the neural network model of network parameter, generation identification character.
In one of the embodiments, it is described by the often row target pixel points and and fixed row target pixel points
Handled with importing in the neural network model comprising initial neural network parameter, generation identification character, including：
By the often row target pixel points with and fixed row target pixel points with the feature of generation corresponding dimension to
Amount；
Neural network model of the characteristic vector input comprising initial neural network parameter is handled, generation is known
Malapropism accords with.
In one of the embodiments, the neural network model includes feedback neural network model；The nerve net
Network parameter includes the number of hidden layer node in the feedback neural network model.
A kind of method for recognizing verification code, methods described include：
Obtain identifying code to be known；
The identifying code to be identified is converted into the monocase identifying code of respective amount, included in each monocase identifying code
One character to be identified；
Extract the character feature data in each monocase identifying code；
The character feature data are imported in the construction method of the neural network model for identifying code identification and appointed
Handled in the neural network model gone out constructed by one, generate the single character identified；
According to character corresponding to identifying code to be known described in each single character generation.
In one of the embodiments, the monocase that the identifying code to be identified is converted into respective amount is verified
Code, including：
The identifying code to be identified is generated into two-value identifying code by image preprocessing；
The two-value identifying code is divided into the monocase identifying code of respective amount.
A kind of construction device of neural network model for identifying code identification, described device include：
Identifying code acquisition module is trained, for obtaining multiple training identifying codes, each train is treated in identifying code comprising one
Identify character；
Character feature data extraction module, for extracting the character feature data in each training identifying code；
Character generation module is identified, initial god is included for the character feature data in the training identifying code to be imported
Handled in neural network model through network parameter, generation identification character；
Identification error rate computing module, for being calculated according to each identification character with the checking character of corresponding training identifying code
The identification error rate of the neural network model；
Neural network model builds module, for when the identification error rate is more than error-rate threshold, according to the knowledge
Malapropism accords with the neural network parameter that the neural network model is adjusted with the training identifying code, and the neutral net after adjustment is joined
Number returns as initial neural network parameter and imports the character feature data in the training identifying code comprising cook
The step of being handled in the neural network model of neural network parameter, generating identification character continues to train, until the identification
Error rate is less than error-rate threshold；Neural network parameter after last time is adjusted is as constructed neural network model
Neural network parameter.
In one of the embodiments, described device includes：
Monocase identifying code acquisition module, for obtaining identifying code to be known, the identifying code to be identified is converted into correspondingly
The monocase identifying code of quantity, a character to be verified is included in each monocase identifying code；
Character feature data extraction module, for extracting the character feature data in each monocase identifying code；
Character feature data identification module, for the character feature data to be imported into the god for identifying code identification
Handled in the neural network model gone out constructed by any one of construction method through network model, generation is identified single
Character；
Identifying code character generation module, for the character corresponding to identifying code to be known according to each single character generation.
A kind of computer-readable recording medium, is stored thereon with computer program, and the program is realized when being executed by processor
The step of construction method of the neural network model for being used for identifying code identification described in above-mentioned each embodiment.
A kind of computer-readable recording medium, is stored thereon with computer program, and the program is realized when being executed by processor
In above-mentioned each embodiment the step of method for recognizing verification code.
A kind of computer equipment, including memory, processor and storage can be run on a memory and on a processor
Computer program, realize during the computing device described program and be used for identifying code identification described in above-mentioned each embodiment
The step of construction method of neural network model.
A kind of computer equipment, including memory, processor and storage can be run on a memory and on a processor
Computer program, method for recognizing verification code described in above-mentioned each embodiment is realized during the computing device described program
Step.
Construction method, device, storage medium and the computer equipment of the above-mentioned neural network model for identifying code identification,
When being trained to neural network model, using identify character with it is corresponding train identifying code checking character as feedback adjustment according to
According to neural network model is adjusted, to build the neutral net mould that can be used for identifying code identification of the error rate less than error-rate threshold
Type.Wherein, it is defeated by the way that the character feature data in multiple training identifying codes are used as when being trained to neural network model
Enter, by being handled in the neural network model comprising initial neural network parameter, generation identification character, so as to calculate
To identification error rate.Neural network model is built by optimization neural network parameter and known to identify that identifying code can improve identifying code
Other accuracy, the time of reduction identification identifying code, so as to improve the efficiency of identifying code identification.
Method, apparatus, storage medium and the computer equipment of above-mentioned identifying code identification, the identifying code to be known of acquisition is changed
Into the monocase identifying code of respective amount, the character feature data of each monocase identifying code are imported to the nerve net built in advance
Network model carries out single character corresponding to processing generation, the character according to corresponding to each single character generates identifying code to be known, and leads to
The neural network parameter for crossing structure builds neural network model to identify that identifying code can improve the accuracy of identifying code identification, contracting
Subtract the time of identification identifying code, so as to improve the efficiency of identifying code identification.
Brief description of the drawings
Fig. 1 is the internal structure schematic diagram of electronic equipment in one embodiment；
Fig. 2 is the flow chart of the construction method for the neural network model for being used for identifying code identification in one embodiment；
Fig. 3 is the flow chart of generation identification character in another embodiment；
Fig. 4 is the schematic diagram of feedback neural network model in one embodiment；
Fig. 5 is the flow chart of method for recognizing verification code in one embodiment；
Fig. 6 is the block diagram of the construction device for the neural network model for being used for identifying code identification in one embodiment；
Fig. 7 is the block diagram that code recognition device is verified in one embodiment.
Embodiment
In order to make the purpose , technical scheme and advantage of the present invention be clearer, it is right below in conjunction with drawings and Examples
The present invention is further elaborated.It should be appreciated that specific embodiment described herein is only to explain the present invention, not
For limiting the present invention.
Fig. 1 is the internal structure schematic diagram of electronic equipment in one embodiment.Reference picture 1, the electronic equipment include passing through
Processor, non-volatile memory medium, built-in storage and the display screen of system bus connection.Wherein, the processing of the electronic equipment
Device is used to provide calculating and control ability, supports the operation of whole electronic equipment.The non-volatile memory medium of the electronic equipment
Operating system and computer executable instructions are stored with, the computer executable instructions can be performed by processor, for reality
A kind of construction method for neural network model for identifying code identification that existing following embodiment provides.The built-in storage is non-
Operating system, computer executable instructions in volatile storage medium provide the running environment of cache.The electronic equipment
Can be terminal or server.Terminal can be that personal computer obtains at least one of mobile electronic device etc..
Server can be realized with the server cluster that the either multiple physical servers of independent server form.
It will be understood by those skilled in the art that the structure shown in Fig. 1, the only part related to application scheme knot
The block diagram of structure, does not form the restriction for the electronic equipment being applied thereon to application scheme, and specific electronic equipment can be with
Including than more or less parts shown in figure, either combining some parts or being arranged with different parts.Such as
The electronic equipment may also include network interface, for carrying out network service with other electronic equipments, obtain identifying code.
In another embodiment, the non-volatile memory medium of the electronic equipment shown in Fig. 1 be stored with operating system and
Computer executable instructions, the computer executable instructions are also with realizing that a kind of identifying code provided in the embodiment of the present invention is known
Other method.
As shown in Fig. 2 in one embodiment, there is provided a kind of structure side of neural network model for identifying code identification
Method, comprise the following steps：
Step S202, multiple training identifying codes are obtained, each train in identifying code includes a character to be identified.
In the present embodiment, training identifying code refers to the monocase identifying code picture for training neutral net, each training
A character to be identified is included in identifying code.Wherein, character to be identified includes but is not limited to numeral and English alphabet etc..Specifically
Ground, multiple corresponding training identifying codes of identifying code picture generation can be downloaded from website, for example be preserved by way of sectional drawing
Identifying code picture, or downloaded and verified according to the URL (Uniform Resource Locator, URL) of website
Code picture, by the multiple corresponding training identifying codes of multiple identifying code pictures generation of acquisition.Further, image preprocessing can be passed through
Identifying code picture is generated into training identifying code, such as, picture is subjected to the one of which or more such as gray processing, binaryzation and denoising
The identifying code picture of kind processing mode generation binaryzation, carries out Character segmentation, generation is corresponding by the identifying code picture of the binaryzation
The training identifying code of quantity.Wherein, the identifying code picture of the binaryzation includes comprising only the checking of black pixel and white pixel point
Code picture.In one embodiment, operation generation identifying code picture can also be made by receiving corresponding identifying code, wherein,
The identifying code picture of generation has same disturbance feature, including but not limited to interfering line, word with the identifying code picture for needing to identify
The combination of the one or more of which such as adhesion or character distortion is accorded with, and the identifying code picture of generation is subjected to image preprocessing and word
Symbol segmentation obtains training identifying code.In one embodiment, the training identifying code for including a character to be identified can be directly generated,
Save the training identifying code generation time.
In one embodiment, the identifying code picture of acquisition can be different resolution, then image preprocessing may also include pair
Image is normalized, the operation such as scaling, translation by the way that identifying code picture to be carried out to isomorphism yardstick, by each identifying code
Picture is all mapped in generation training identifying code in unified canonical matrix.For example, unified canonical matrix can be for 15*20's
Matrix.Identifying code image is normalized, it is ensured that training identifying code has uniformity, convenient to neural network model
It is trained.
In one embodiment, identifying code picture after image preprocessing is subjected to Character segmentation, generates the instruction of respective amount
Practice identifying code, including identifying code picture is split according to connected region partitioning algorithm.Specifically, enter to identifying code picture
Before row segmentation, in addition to corrosion expansive working is carried out to identifying code picture, can be first to corrode the closed operation expanded afterwards, or
First expand post-etching opens operation.By corroding expansive working, character can be refined, removes interfering line so as to identifying code figure
The segmentation of piece is more accurate.
In one embodiment, the character to be identified that can be included according to training identifying code is classified.For example if wait to know
Malapropism symbol includes 1 to 9 nine numeral, can regard each numeral as a kind of training identifying code, identifying code picture is pre-processed
And the training identifying code after segmentation is classified according to the character to be identified included.Identifying code will be trained according to classification to nerve
Network model is trained, and can improve the efficiency of training neural network model.
Step S204, extract the character feature data in each training identifying code.
In the present embodiment, character feature data refer to the feature that can be identified for that the character to be identified included in training identifying code
Data.Specifically, characteristic includes but is not limited to the spy that the pixel that character to be identified occupies in identifying code is trained has
Levy data.For example, characteristic can be character to be identified it is occupied pixel in training identifying code is often gone and, treat
Identify every fritter of character pixel the and/or to be identified character occupied in identifying code each column is trained in training identifying code
It is occupied pixel and wait in region, but not limited to this.Further, can be extracted by MATLAB (a kind of mathematical software)
Character feature data in each training identifying code.
Step S206, the character feature data in identifying code will be trained to import the nerve for including initial neural network parameter
Handled in network model, generation identification character.
In the present embodiment, neural network model refers to the complex network model being interconnected to form by multilayer, species bag
Include but be not limited to feedforward neural network, Feedback Neural Network and self organizing neural network etc..Identification character refers to by nerve net
The character included in the training identifying code that network Model Identification goes out, accordingly, include but is not limited to numeral and English alphabet etc..Nerve
Network parameter refers to used parameter in the component of neural network model, including but not limited to neuron number and connection weight
Value etc..One neural network model comprising initial neural network parameter can be trained, the god that training is built after terminating
Identified through network model available for identifying code.Neural network model may include multilayer neural network layer, every layer of neural net layer it
Between be all attached by connection weight, the neural network parameter in neural network model can be multiple, every layer of neutral net
Layer receives the operation result of preceding layer, by the computing of itself, to the next layer of operation result for exporting this layer.For example, it is refreshing
May include input layer, hidden layer and output layer through network model, each neuron node and input layer in hidden layer it is each
Input block is connected by connection weight, and the neuron node of output layer is connected with the neuron node in hidden layer.
Specifically, after the character feature data in each training identifying code of extraction, each character feature data are distinguished
Neural network model is inputted as input layer, passes sequentially through each hidden layer of neural network model.In every layer of hidden layer
On, using nonlinear change operator corresponding to the hidden layer, non-linear meter is carried out to the character feature data of last layer output
Calculate, and final output identification character.In one embodiment, neural network model can only include one layer of hidden layer, reduce god
Complexity through network model.
Step S208, neural network model is calculated with the checking character of corresponding training identifying code according to each identification character
Identification error rate.
In the present embodiment, checking character refers to train the character that truly includes in identifying code, training identifying code be by
The identifying code picture generated on the basis of checking character plus disturbing factor, such as interfering line, Character deformation and Characters Stuck
Deng.Identification error rate refers to by the way that relatively each whether identification character is consistent with the checking character of corresponding training identifying code and calculating
The probability of generation, the identification error rate fall within neural network parameter.For example 100 training are identified by neural network model
Identifying code, wherein there is 70, then nerve nets consistent with the checking character of corresponding training identifying code in the identification character identified
The identification error rate of network model is 30%.
Step S210, when identification error rate is more than error-rate threshold, according to identification character and training identifying code adjustment god
Neural network parameter through network model, using the neural network parameter after adjustment as initial neural network parameter, and return
Located training the character feature data in identifying code to import in the neural network model comprising initial neural network parameter
Reason, generate the step of identifying character and continue to train, until identification error rate is less than error-rate threshold；After last time is adjusted
Neural network parameter of the neural network parameter as constructed neural network model.
In the present embodiment, the process for training neural network model is neutral net in the neural network model that determination need to be trained
The process of parameter.It is determined that during neural network parameter, electronic equipment can be initialized first in the neural network model that need to be trained
Neural network parameter, and in follow-up training process, continue to optimize the neural network parameter and return in training identifying code
Character feature data import comprising initial neural network parameter neural network model in handled, generation identification character
The step of, until the neural network model being calculated by each identification character with the checking character of corresponding training identifying code
Identification error rate is less than error-rate threshold, after the completion of neural network parameter training, optimizes obtained god after last time is adjusted
Neural network parameter through network parameter as constructed neural network model.
In one embodiment, can also preset to return will train the character feature data in identifying code to import comprising initial
Neural network parameter neural network model in handled, number the step of generation identification character, such as 10 times, 15 times
Or 20 is inferior.When recycle time reaches preset times, stop the training of neutral net, optimize after last time is adjusted and obtain
Neural network parameter of the neural network parameter as constructed neural network model.
In one embodiment, after having built neural network model, test identifying code can also be obtained, test checking
Code can be consistent with training identifying code, or downloads from website or tested by receiving identifying code making operation and what is generated again
Demonstrate,prove code picture.Identifying code will be tested and obtain the monocase test identifying code of respective amount by corresponding image preprocessing and segmentation
Afterwards, the character feature data in monocase test identifying code are extracted.The character that each monocase is tested in identifying code is special
Sign data, which are imported into the neural network model built in advance, to be handled, the knowledge corresponding to generation monocase test identifying code
Malapropism accords with.The test word corresponding to identification character generation test identifying code according to corresponding to each monocase tests identifying code
Symbol, the identification that neural network model is calculated according to the validation test character of the corresponding test identifying code of each test character miss
Rate.When identification error rate is more than error-rate threshold, returning will train the character feature data in identifying code to import comprising just
Handled in the neural network model of the neural network parameter of beginning, generate the step of identifying character, continue to neutral net mould
Type is trained, until the identification error rate of the neural network model by test identifying code test is less than error-rate threshold, and
The neural network parameter of neural network model of the neural network parameter as constructed by after last time is adjusted.
In one embodiment, can be verified the test identifying code tested neural network model as training
Code, for the subsequently training to neural network model.Specifically, the identification error that will can be calculated by neural network model
Rate is higher than the test identifying code corresponding to predetermined threshold value as training identifying code.
The construction method of the above-mentioned neural network model for identifying code identification, is trained to neural network model
When, identification character is adjusted into neural network model with the checking character of corresponding training identifying code as feedback adjustment foundation, with
Build the neural network model that can be used for identifying code identification that error rate is less than error-rate threshold.Wherein, to neutral net mould
When type is trained, by regarding the character feature data in multiple training identifying codes as input, by including initial nerve
Handled in the neural network model of network parameter, generation identification character, so as to which identification error rate be calculated.Pass through optimization
Neural network parameter builds neural network model to identify that identifying code can improve the accuracy of identifying code identification, and reduction identification is tested
The time of code is demonstrate,proved, so as to improve the efficiency of identifying code identification.
In one embodiment, character feature data include every row target pixel points and and fixed row target pixel points
Sum, target pixel points are the pixels that character to be identified occupies in corresponding training identifying code, and step S206 includes：Will
Often row target pixel points and and fixed row target pixel points and import the nerve net for including initial neural network parameter
Handled in network model, generation identification character.
In the present embodiment, training identifying code can be expressed as matrix, the height of the corresponding training identifying code image of row of matrix, square
The width of the corresponding training identifying code image of row of battle array, the element of matrix correspondingly train the pixel of identifying code image, the value of matrix element
It can be the gray value of pixel.Target pixel points and refer to shared by the character to be identified that is drawn by matrix data in analysis matrix
Pixel.Fixed row refer to the row with identification numerical value drawn by the distribution of the sum of each column target pixel points.Citing
For, verifying may include that at least three fixes row, the object pixel of the row at two perpendicular places in the training identifying code that character is alphabetical H
Point and two it is perpendicular between row, when be calculated the fixation row target pixel points and when within preset range, with reference to every row mesh
The distribution of the sum of pixel is marked, by the analyzing and processing of neural network model, generation identification character H.
In one embodiment, character feature data can also be fixed row target pixel points and and each column target picture
The sum of vegetarian refreshments, or fixed row target pixel points and and fixed row target pixel points sum.
In one embodiment, it is pre- by one or more of which such as gray processing, binaryzation and denoisings to train identifying code
Handle and the two-value identifying code picture of generation, black pixel and white pixel point are included in the two-value identifying code picture, if to be identified
The pixel that character occupies in corresponding training identifying code is black pixel, and background is white pixel point, then the training is verified
Often row target pixel points and for often capable black pixel the quantity sums of the character feature data of code, fixed row target pixel points
For the quantity sum of the black pixel of fixed row.It is possible to further which the matrix element value where black pixel in matrix is set
Be set to 1, the matrix element value where white pixel point is arranged to 0, then often row target pixel points with 1 to be added in often being gone in matrix
Sum, fixed row target pixel points with matrix it is fixed arrange in 1 sum being added.
In one embodiment, as shown in figure 3, by every row target pixel points and and fixed row target pixel points sum
Import in the neural network model comprising initial neural network parameter and handled, generation identification character specifically includes following step
Suddenly：
Step S302, by every row target pixel points with and fixed row target pixel points spy with the corresponding dimension of generation
Sign vector.
In the present embodiment, characteristic vector refers to there is corresponding latitude according to what the quantity of training identifying code row and column was generated
Vector, value in each dimension of this feature vector and often row pixel and and each column target pixel points and one a pair
Should.For example, pixel matrix one shares 20 rows and 15 row, then the character feature data of the training identifying code can generate 35 dimensions
The characteristic vector of degree, wherein, except the value in the corresponding dimension of other row of fixed row can be 0.
Step S304, neural network model of the characteristic vector input comprising initial neural network parameter is handled,
Generation identification character.
In the present embodiment, identification character refers to be identified by the neural network model comprising initial neural network parameter
The character included in the training identifying code gone out.Using the characteristic vector of corresponding dimension as the god for including initial neural network parameter
Input neuron through network model, the identification character identified by the calculating and processing generation of neural network model.Wherein,
The quantity for inputting neuron is corresponding with the number of dimensions of characteristic vector.For example, the characteristic vector of one 35 dimension is inputted
Neural network model is handled, then the input layer of the neural network model includes 35 neurons.Specifically, the god of input layer
It is attached through member by the hidden layer of connection weight and neural network model, output layer can be obtained by last layer of hidden layer
Output.Wherein, output layer can include a neuron, and the neuron includes the identification character of generation.
In one embodiment, neural network model can adjust input layer nerve according to the dimension of the characteristic vector of input
The quantity of member, so as to reduce the limitation of identification identifying code so that neural network model can identify more kinds of checkings
Code.
In one embodiment, the output layer of neural network model can be possible to character for the training identifying code is corresponding
Probability.Such as being possible to character includes 9 numerals, 26 English alphabets, then the neuron of output layer is 35, respectively
It is the probability that the midamble code is each character.
In one embodiment, neural network model includes feedback neural network model；Neural network parameter includes anti-
The number of hidden layer node in feedback type neural network model.
In the present embodiment, the schematic diagram of the feedback neural network model provided with reference to figure 4, the feedback neural network mould
Type includes input layer 402, hidden layer 404 and output layer 406, wherein, multiple input layer sections are included in input layer 402
Point, multiple hidden layer nodes are included in hidden layer 404, and hidden layer 404 can be multilayer, output layer 406 can include an output
Layer neuron node.Feedback neural network model refers to that the output layer 406 in the neural network model has instead with input layer 402
Feedback connection.For example, the output layer 406 of neural network model can be the word for the optimization being calculated by optimizing connection weight
Characteristic, such as the characteristic vector with optimization dimension are accorded with, can be using the character feature data after optimization as neutral net
Input neuron in the input layer 402 of model is calculated and handled again.Specifically, can be by adjusting feedback-type nerve
The number of hidden layer node carrys out the output of optimization neural network model in network.
In one embodiment, as shown in Figure 5, there is provided a kind of method for recognizing verification code, this method include：
Step S502, obtain identifying code to be known.
In the present embodiment, identifying code to be known refers to the identifying code picture that needs are identified, and the identifying code to be known can pass through
The URL of webpage comprising identifying code to be known is downloaded acquisition, or can also be shown on a display screen by directly scanning
Identifying code picture is obtained.
Step S504, identifying code to be identified is converted into the monocase identifying code of respective amount, each monocase identifying code
In include a character to be identified.
In the present embodiment, multiple monocase identifying codes are included in identifying code to be identified, can be incited somebody to action according to default conversion regime
The monocase identifying code of identifying code conversion generation respective amount to be identified, for example, if the word included in identifying code to be identified
Accord with as 4H7T, the identifying code to be identified can be converted to four monocase identifying codes, be 4, H, 7, T list respectively comprising character
Character identifying code.
Step S506, extract the character feature data in each monocase identifying code.
In the present embodiment, character feature data refer to the spy that can be identified for that the character to be identified included in monocase identifying code
Levy data.Specifically, characteristic, which includes but is not limited to the pixel that character to be identified occupies in monocase identifying code, has
Characteristic.For example, characteristic can be character to be identified pixel occupied in monocase identifying code is often gone
And, character to be identified pixel and/or to be identified character occupied in monocase identifying code each column verifies in monocase
Pixel occupied by every pocket of code and wait, but not limited to this.Further, can be extracted by MATLAB each
Character feature data in monocase identifying code.
Step S508, character feature data are imported in the neural network model constructed in advance and handled, generate institute
The single character identified.
In the present embodiment, the neural network model constructed in advance is to be known by being used for identifying code in above-mentioned each embodiment
The construction method of other neural network model and the neural network model built.Specifically, character feature data may include every row
Target pixel points and and fixed row pixel sum.By each monocase identifying code often row target pixel points and it is and solid
Determine neural network model row pixel and that importing is constructed in advance to be handled.It is possible to further be tested according to monocase
Demonstrate,prove code often row target pixel points and and fixation row pixel the characteristic vector with the corresponding dimension of generation, characteristic vector is defeated
Enter the neural network model constructed in advance to be handled, generate the single character identified.
In one embodiment, black pixel and white pixel point is only included in monocase identifying code, and treats character learning symbol at this
The pixel occupied in monocase identifying code is black pixel.Can extract the monocase identifying code in often go black pixel and with
And fixed sum for arranging black pixel, and by the black pixel of every row and and fixation arrange black pixel with the corresponding dimension of generation
Characteristic vector, the input layer using this feature vector as neural network model, passes through the identification of neutral net, generation pair
The single character answered.
Step S510, the character according to corresponding to each single character generates identifying code to be known.
In the present embodiment, the neutral net that the character feature data input of each monocase identifying code can be built in advance
In model, the single character of the identified respective numbers of generation, according to the single character of the respective numbers according to character sequence
Character corresponding to identifying code to be known can be generated.For example, if each single character of generation is respectively 4, H, 7, T in order,
Then character corresponding to identifying code to be known is 4H7T.
Above-mentioned identifying code knows method for distinguishing, and the identifying code to be known of acquisition is converted into the monocase identifying code of respective amount,
The character feature data of each monocase identifying code are imported to the neural network model built in advance to carry out corresponding to processing generation
Single character, the character according to corresponding to each single character generates identifying code to be known, built by the neural network parameter of structure
Neural network model come identify identifying code can improve identifying code identification accuracy, reduction identification identifying code time, so as to
Improve the efficiency of identifying code identification.
In one embodiment, step S504 includes：Identifying code to be identified is generated into two-value by image preprocessing to verify
Code；Two-value identifying code is divided into the monocase identifying code of respective amount.
In the present embodiment, two-value identifying code refers to can be used in identifying code to be known by image preprocessing conversion generation
The identifying code picture of image feature data is extracted, the identifying code picture can be the identifying code picture of binaryzation.Wherein, image is located in advance
Reason includes but is not limited to the combination image processing mode of the one or more of which such as gray processing, binaryzation and denoising.Specifically, two
Value identifying code can only include black pixel and white pixel point.Further, the two-value identifying code is subjected to Character segmentation, generation
The monocase identifying code of respective amount.Wherein, partitioning scheme can be connected region split plot design.Specifically, to identifying code picture
Before being split, in addition to corrosion expansive working is carried out to identifying code picture, can be first to corrode the closed operation expanded afterwards, also may be used
Operation is opened for first expand post-etching.By image preprocessing and corrosion expansive working, character can be refined, removes interfering line,
So that the segmentation to identifying code picture is more accurate, the monocase identifying code acquired is easier identified.
In one embodiment, as shown in Figure 6, there is provided a kind of structure dress of neural network model for identifying code identification
Put, it is characterised in that the device includes：
Identifying code acquisition module 602 is trained, for obtaining multiple training identifying codes, each train includes one in identifying code
Character to be identified.
Character feature data extraction module 604, for extracting the character feature data in each training identifying code.
Character generation module 606 being identified, initial god is included for the character feature data in identifying code will be trained to import
Handled in neural network model through network parameter, generation identification character.
Identification error rate computing module 608, for the checking character according to each identification character with corresponding training identifying code
Calculate the identification error rate of neural network model.
Neural network model builds module 610, for when identification error rate is more than error-rate threshold, according to identifying character
The neural network parameter of neural network model is adjusted with training identifying code, using the neural network parameter after adjustment as initial god
Through network parameter, and return to the nerve that the character feature data in identifying code will be trained to import the neural network parameter comprising cook
The step of being handled in network model, generating identification character continues to train, until identification error rate is less than error-rate threshold；Will
Neural network parameter of the neural network parameter as constructed neural network model after last time adjustment.
In one embodiment, character feature data include every row target pixel points and and fixed row target pixel points
Sum, target pixel points are the pixels that character to be identified occupies in corresponding training identifying code；Identify character generation mould
Block 606 be additionally operable to it is by every row target pixel points and and fixed row target pixel points and import and include initial neutral net
Handled in the neural network model of parameter, generation identification character.
In one embodiment, identification character generation module 606 is additionally operable to by every row target pixel points and and fixed
The characteristic vector with the corresponding dimension of generation of row target pixel points, by characteristic vector input comprising initial neural network parameter
Neural network model is handled, generation identification character.
In one embodiment, neural network model includes feedback neural network model；Neural network parameter includes anti-
The number of hidden layer node in feedback type neural network model.
In one embodiment, as shown in Figure 7, there is provided one kind checking code recognition device, it is characterised in that the device bag
Include：
Monocase identifying code acquisition module 702, for obtaining identifying code to be known, identifying code to be identified is converted into corresponding number
The monocase identifying code of amount, a character to be verified is included in each monocase identifying code.
Character feature data extraction module 704, for extracting the character feature data in each monocase identifying code.
Character feature data identification module 706, for character feature data to be imported into the neutral net mould constructed in advance
Handled in type, generate the single character identified.
Identifying code character generation module 708, for generating character corresponding to identifying code to be known according to each single character.
In one embodiment, monocase identifying code acquisition module 702 is additionally operable to identifying code to be identified is pre- by image
Processing generation two-value identifying code, two-value identifying code is divided into the monocase identifying code of respective amount.
In one embodiment, there is provided a kind of computer-readable recording medium, be stored thereon with computer program, the journey
Following steps are realized when sequence is executed by processor:Multiple training identifying codes are obtained, each train waits to know in identifying code comprising one
Malapropism accords with；Character feature data in each training identifying code of extraction；The character feature data in identifying code will be trained to import bag
Handled in neural network model containing initial neural network parameter, generation identification character；According to it is each identification character with
The checking character of corresponding training identifying code calculates the identification error rate of neural network model；When identification error rate is more than error rate threshold
During value, according to identification character and the neural network parameter for training identifying code adjustment neural network model, by the nerve net after adjustment
Network parameter is as initial neural network parameter, and returning will train the character feature data importing in identifying code to include initially
The step of being handled in the neural network model of neural network parameter, generating identification character continues to train, until identification error
Rate is less than error-rate threshold；The nerve of neural network model of the neural network parameter as constructed by after last time is adjusted
Network parameter.
In one embodiment, character feature data include every row target pixel points and and fixed row target pixel points
Sum, target pixel points are the pixels that character to be identified occupies in corresponding training identifying code, and the program is by processor
During execution, that is realized will train the character feature data in identifying code to import the nerve net for including initial neural network parameter
Handled in network model, generate the step of identifying character, including：By every row target pixel points and and fixed row target picture
Handled in neural network model vegetarian refreshments and that importing includes initial neural network parameter, generation identification character.
In one embodiment, when the program is executed by processor, realized by every row target pixel points and and
Handled in neural network model fixed row target pixel points and that importing includes initial neural network parameter, generation is known
The step of malapropism accords with, specifically includes following steps：By every row target pixel points and and fixed row target pixel points and it is raw
Into the characteristic vector of corresponding dimension；At neural network model of the characteristic vector input comprising initial neural network parameter
Reason, generation identification character.
In one embodiment, neural network model includes feedback neural network model；Neural network parameter includes anti-
The number of hidden layer node in feedback type neural network model.
In one embodiment, there is provided a kind of computer-readable recording medium, be stored thereon with computer program, the journey
Following steps are realized when sequence is executed by processor:Obtain identifying code to be known；Identifying code to be identified is converted into the list of respective amount
Character identifying code, a character to be identified is included in each monocase identifying code；Extract the character in each monocase identifying code
Characteristic；Character feature data are imported to the structure for the neural network model for being used for identifying code identification that above-described embodiment provides
Handled in the neural network model gone out constructed by any one of method, generate the single character identified；According to each
Single character generates character corresponding to identifying code to be known.
In one embodiment, when the program is executed by processor, that is realized is converted into identifying code to be identified correspondingly
The step of monocase identifying code of quantity, specifically include following steps：Identifying code to be identified is generated two by image preprocessing
It is worth identifying code；Two-value identifying code is divided into the monocase identifying code of respective amount.
In one embodiment, a kind of computer equipment, including memory, processor and storage are on a memory and can be
The computer program run on processor, processor realize following steps in configuration processor:Multiple training identifying codes are obtained, often
A character to be identified is included in individual training identifying code；Character feature data in each training identifying code of extraction；Training is tested
Character feature data in card code, which are imported in the neural network model comprising initial neural network parameter, to be handled, and generation is known
Malapropism accords with；The identification error of neural network model is calculated with the checking character of corresponding training identifying code according to each identification character
Rate；When identification error rate is more than error-rate threshold, according to identification character and the god for training identifying code adjustment neural network model
Through network parameter, using the neural network parameter after adjustment as initial neural network parameter, and return in training identifying code
Character feature data import comprising initial neural network parameter neural network model in handled, generation identification character
The step of continue to train, until identification error rate is less than error-rate threshold；Neural network parameter after last time is adjusted is made
For the neural network parameter of constructed neural network model.
In one embodiment, character feature data include every row target pixel points and and fixed row target pixel points
Sum, target pixel points are the pixels that character to be identified occupies in corresponding training identifying code, and above-mentioned processor is held
During line program, that is realized will train the character feature data in identifying code to import the nerve for including initial neural network parameter
Handled in network model, generate the step of identifying character, including：By every row target pixel points and and fixed row target
Handled in neural network model pixel and that importing includes initial neural network parameter, generation identification character.
In one embodiment, during above-mentioned computing device program, realized by every row target pixel points and with
And handled in the neural network model for including initial neural network parameter with importing of fixed row target pixel points, generate
The step of identifying character, specifically includes following steps：By every row target pixel points and and fixed row target pixel points sum
The characteristic vector of the corresponding dimension of generation；Neural network model of the characteristic vector input comprising initial neural network parameter is carried out
Processing, generation identification character.
In one embodiment, neural network model includes feedback neural network model；Neural network parameter includes anti-
The number of hidden layer node in feedback type neural network model.
In one embodiment, a kind of computer equipment, including memory, processor and storage are on a memory and can be
The computer program run on processor, processor realize following steps in configuration processor:Obtain identifying code to be known；It will wait to know
Other identifying code is converted into the monocase identifying code of respective amount, and a character to be identified is included in each monocase identifying code；Carry
Take the character feature data in each monocase identifying code；It is used to verify by what character feature data imported that above-described embodiment provides
Handled in the neural network model gone out constructed by any one of construction method of neural network model of code identification, generate institute
The single character identified；The character according to corresponding to each single character generates identifying code to be known.
In one embodiment, during above-mentioned computing device program, that is realized is converted into identifying code to be identified pair
The step of answering the monocase identifying code of quantity, specifically includes following steps：Identifying code to be identified is generated by image preprocessing
Two-value identifying code；Two-value identifying code is divided into the monocase identifying code of respective amount.
One of ordinary skill in the art will appreciate that realize all or part of flow in above-described embodiment method, being can be with
The hardware of correlation is instructed to complete by computer program, described program can be stored in a non-volatile computer and can be read
In storage medium, the program is upon execution, it may include such as the flow of the embodiment of above-mentioned each method.Wherein, described storage is situated between
Matter can be magnetic disc, CD, read-only memory (Read-Only Memory, ROM) etc..
Each technical characteristic of embodiment described above can be combined arbitrarily, to make description succinct, not to above-mentioned reality
Apply all possible combination of each technical characteristic in example to be all described, as long as however, the combination of these technical characteristics is not deposited
In contradiction, the scope that this specification is recorded all is considered to be.
Embodiment described above only expresses the several embodiments of the present invention, and its description is more specific and detailed, but simultaneously
Can not therefore it be construed as limiting the scope of the patent.It should be pointed out that come for one of ordinary skill in the art
Say, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to the protection of the present invention
Scope.Therefore, the protection domain of patent of the present invention should be determined by the appended claims.
Claims (10)
1. a kind of construction method of neural network model for identifying code identification, methods described include：
Multiple training identifying codes are obtained, each train in identifying code includes a character to be identified；
Character feature data in each training identifying code of extraction；
Character feature data in the training identifying code are imported to the neural network model for including initial neural network parameter
In handled, generation identification character；
The identification error of the neural network model is calculated with the checking character of corresponding training identifying code according to each identification character
Rate；
When the identification error rate is more than error-rate threshold, according to the identification character with the training identifying code adjustment
The neural network parameter of neural network model, using the neural network parameter after adjustment as initial neural network parameter, and return
Return and import the character feature data in the training identifying code in the neural network model comprising initial neural network parameter
The step of being handled, generating identification character continues to train, until the identification error rate is less than error-rate threshold；By last
Neural network parameter of the neural network parameter as constructed neural network model after secondary adjustment.
2. according to the method for claim 1, it is characterised in that the character feature data include often row target pixel points
With and fixed row target pixel points sum, the target pixel points are the characters to be identified in corresponding training identifying code
In the pixel that occupies；
Character feature data in the training identifying code are imported to the neural network model for including initial neural network parameter
In handled, generation identification character, including：
By the often row target pixel points and and fixed row target pixel points and import and join comprising initial neutral net
Handled in several neural network models, generation identification character.
3. according to the method for claim 2, it is characterised in that it is described will the often row target pixel points and it is and fixed
Handled in neural network model row target pixel points and that importing includes initial neural network parameter, generation identification word
Symbol, including：
By the often row target pixel points with and fixed row target pixel points characteristic vector with the corresponding dimension of generation；
Neural network model of the characteristic vector input comprising initial neural network parameter is handled, generation identification word
Symbol.
4. according to the method for claim 1, it is characterised in that the neural network model includes feedback neural network mould
Type；The neural network parameter includes the number of hidden layer node in the feedback neural network model.
5. a kind of method for recognizing verification code, methods described include：
Obtain identifying code to be known；
The identifying code to be identified is converted into the monocase identifying code of respective amount, one is included in each monocase identifying code
Character to be identified；
Extract the character feature data in each monocase identifying code；
The character feature data are imported in the neural network model gone out constructed by any one of Claims 1-4 and located
Reason, generate the single character identified；
According to character corresponding to identifying code to be known described in each single character generation.
6. according to the method for claim 5, it is characterised in that described that the identifying code to be identified is converted into respective amount
Monocase identifying code, including：
The identifying code to be identified is generated into two-value identifying code by image preprocessing；
The two-value identifying code is divided into the monocase identifying code of respective amount.
7. the construction device of a kind of neural network model for identifying code identification, it is characterised in that described device includes：
Identifying code acquisition module is trained, it is to be identified comprising one in each training identifying code for obtaining multiple training identifying codes
Character；
Character feature data extraction module, for extracting the character feature data in each training identifying code；
Character generation module is identified, initial nerve net is included for the character feature data in the training identifying code to be imported
Handled in the neural network model of network parameter, generation identification character；
Identification error rate computing module, described in being calculated according to each identification character with the checking character of corresponding training identifying code
The identification error rate of neural network model；
Neural network model builds module, for when the identification error rate is more than error-rate threshold, according to the identification word
Symbol adjusts the neural network parameter of the neural network model with the training identifying code, and the neural network parameter after adjustment is made
For initial neural network parameter, and return to the nerve that the character feature data in the training identifying code are imported and include cook
The step of being handled in the neural network model of network parameter, generating identification character continues to train, until the identification error
Rate is less than error-rate threshold；The nerve of neural network model of the neural network parameter as constructed by after last time is adjusted
Network parameter.
8. one kind checking code recognition device, it is characterised in that described device includes：
Monocase identifying code acquisition module, for obtaining identifying code to be known, the identifying code to be identified is converted into respective amount
Monocase identifying code, include a character to be verified in each monocase identifying code；
Character feature data extraction module, for extracting the character feature data in each monocase identifying code；
Character feature data identification module, for the character feature data to be imported constructed by any one of Claims 1-4
Handled in the neural network model gone out, generate the single character identified；
Identifying code character generation module, for the character corresponding to identifying code to be known according to each single character generation.
9. a kind of computer-readable recording medium, is stored thereon with computer program, it is characterised in that the program is held by processor
The step of any one methods described in claim 1 to 6 is realized during row.
10. a kind of computer equipment, including memory, processor and storage are on a memory and the meter that can run on a processor
Calculation machine program, it is characterised in that side described in any one in claim 1 to 6 is realized during the computing device described program
The step of method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710453765.4A CN107360137A (en) | 2017-06-15 | 2017-06-15 | Construction method and device for the neural network model of identifying code identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710453765.4A CN107360137A (en) | 2017-06-15 | 2017-06-15 | Construction method and device for the neural network model of identifying code identification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107360137A true CN107360137A (en) | 2017-11-17 |
Family
ID=60273601
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710453765.4A Pending CN107360137A (en) | 2017-06-15 | 2017-06-15 | Construction method and device for the neural network model of identifying code identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107360137A (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109063456A (en) * | 2018-08-02 | 2018-12-21 | 浙江大学 | The safety detecting method and system of image-type identifying code |
| CN109257168A (en) * | 2018-09-21 | 2019-01-22 | 深圳市元征科技股份有限公司 | A kind of generation method of picture validation code, device, equipment and medium |
| CN109284600A (en) * | 2018-08-08 | 2019-01-29 | 贵阳朗玛信息技术股份有限公司 | A kind of method and device of identifying code for identification |
| CN109684815A (en) * | 2018-12-13 | 2019-04-26 | 珍岛信息技术（上海）股份有限公司 | It is a kind of can self-teaching method for recognizing verification code |
| CN109858542A (en) * | 2019-01-25 | 2019-06-07 | 广州云测信息技术有限公司 | A kind of character identifying method and device |
| CN109933969A (en) * | 2017-12-15 | 2019-06-25 | 腾讯科技（深圳）有限公司 | Method for recognizing verification code, device, electronic equipment and readable storage medium storing program for executing |
| CN110348438A (en) * | 2019-06-29 | 2019-10-18 | 上海淇馥信息技术有限公司 | A kind of picture character identifying method, device and electronic equipment based on artificial nerve network model |
| CN110351094A (en) * | 2019-07-03 | 2019-10-18 | 平安科技（深圳）有限公司 | Character verification method, device, computer equipment and storage medium |
| CN110555298A (en) * | 2019-08-30 | 2019-12-10 | 阿里巴巴（中国）有限公司 | Verification code recognition model training and recognition method, medium, device and computing equipment |
| CN110909807A (en) * | 2019-11-26 | 2020-03-24 | 深圳市信联征信有限公司 | Network verification code identification method and device based on deep learning and computer 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 |
| CN111274957A (en) * | 2020-01-20 | 2020-06-12 | 阳光人寿保险股份有限公司 | Webpage verification code identification method, device, terminal and computer storage medium |
| CN111753281A (en) * | 2020-06-30 | 2020-10-09 | 北京鼎泰智源科技有限公司 | Verification code identification method |
| WO2020215573A1 (en) * | 2019-04-22 | 2020-10-29 | 平安科技（深圳）有限公司 | Captcha identification method and apparatus, and computer device and storage medium |
| CN112270325A (en) * | 2020-11-09 | 2021-01-26 | 携程旅游网络技术（上海）有限公司 | Character verification code recognition model training method, recognition method, system, device and medium |
| WO2021027113A1 (en) * | 2019-08-14 | 2021-02-18 | 平安科技（深圳）有限公司 | Image verification method, device, computer apparatus, and storage medium |
| CN114817937A (en) * | 2021-01-19 | 2022-07-29 | 北京嘀嘀无限科技发展有限公司 | Keyboard encryption method, device, storage medium and computer program product |
| CN114817893A (en) * | 2021-01-19 | 2022-07-29 | 北京嘀嘀无限科技发展有限公司 | Authentication code image encryption method, device, storage medium and computer program product |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104021376A (en) * | 2014-06-05 | 2014-09-03 | 北京乐动卓越科技有限公司 | Verification code identifying method and device |
| CN105205448A (en) * | 2015-08-11 | 2015-12-30 | 中国科学院自动化研究所 | Character recognition model training method based on deep learning and recognition method thereof |
| CN105354574A (en) * | 2015-12-04 | 2016-02-24 | 山东博昂信息科技有限公司 | Vehicle number recognition method and device |
| CN106067019A (en) * | 2016-05-27 | 2016-11-02 | 北京旷视科技有限公司 | The method and device of Text region is carried out for image |
| CN106446897A (en) * | 2016-09-09 | 2017-02-22 | 浪潮软件股份有限公司 | Hollow verification code identification method |
-
2017
- 2017-06-15 CN CN201710453765.4A patent/CN107360137A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104021376A (en) * | 2014-06-05 | 2014-09-03 | 北京乐动卓越科技有限公司 | Verification code identifying method and device |
| CN105205448A (en) * | 2015-08-11 | 2015-12-30 | 中国科学院自动化研究所 | Character recognition model training method based on deep learning and recognition method thereof |
| CN105354574A (en) * | 2015-12-04 | 2016-02-24 | 山东博昂信息科技有限公司 | Vehicle number recognition method and device |
| CN106067019A (en) * | 2016-05-27 | 2016-11-02 | 北京旷视科技有限公司 | The method and device of Text region is carried out for image |
| CN106446897A (en) * | 2016-09-09 | 2017-02-22 | 浪潮软件股份有限公司 | Hollow verification code identification method |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109933969A (en) * | 2017-12-15 | 2019-06-25 | 腾讯科技（深圳）有限公司 | Method for recognizing verification code, device, electronic equipment and readable storage medium storing program for executing |
| CN109933969B (en) * | 2017-12-15 | 2021-05-07 | 腾讯科技（深圳）有限公司 | Verification code identification method and device, electronic equipment and readable storage medium |
| CN109063456B (en) * | 2018-08-02 | 2021-10-08 | 浙江大学 | Security detection method and system for image type verification code |
| CN109063456A (en) * | 2018-08-02 | 2018-12-21 | 浙江大学 | The safety detecting method and system of image-type identifying code |
| CN109284600A (en) * | 2018-08-08 | 2019-01-29 | 贵阳朗玛信息技术股份有限公司 | A kind of method and device of identifying code for identification |
| CN109257168A (en) * | 2018-09-21 | 2019-01-22 | 深圳市元征科技股份有限公司 | A kind of generation method of picture validation code, device, equipment and medium |
| CN109684815A (en) * | 2018-12-13 | 2019-04-26 | 珍岛信息技术（上海）股份有限公司 | It is a kind of can self-teaching method for recognizing verification code |
| CN109858542A (en) * | 2019-01-25 | 2019-06-07 | 广州云测信息技术有限公司 | A kind of character identifying method and device |
| WO2020215573A1 (en) * | 2019-04-22 | 2020-10-29 | 平安科技（深圳）有限公司 | Captcha identification method and apparatus, and computer device and storage medium |
| CN110348438A (en) * | 2019-06-29 | 2019-10-18 | 上海淇馥信息技术有限公司 | A kind of picture character identifying method, device and electronic equipment based on artificial nerve network model |
| CN110351094A (en) * | 2019-07-03 | 2019-10-18 | 平安科技（深圳）有限公司 | Character verification method, device, computer equipment and storage medium |
| CN110351094B (en) * | 2019-07-03 | 2024-03-15 | 平安科技（深圳）有限公司 | Character verification method, device, computer equipment and storage medium |
| WO2021027113A1 (en) * | 2019-08-14 | 2021-02-18 | 平安科技（深圳）有限公司 | Image verification method, device, computer apparatus, and storage medium |
| CN110555298B (en) * | 2019-08-30 | 2021-10-26 | 阿里巴巴（中国）有限公司 | Verification code recognition model training and recognition method, medium, device and computing equipment |
| CN110555298A (en) * | 2019-08-30 | 2019-12-10 | 阿里巴巴（中国）有限公司 | Verification code recognition model training and recognition method, medium, device and computing equipment |
| CN110909807A (en) * | 2019-11-26 | 2020-03-24 | 深圳市信联征信有限公司 | Network verification code identification method and device based on deep learning and computer equipment |
| CN111274957A (en) * | 2020-01-20 | 2020-06-12 | 阳光人寿保险股份有限公司 | Webpage verification code identification method, device, terminal and computer 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 |
| CN111753281A (en) * | 2020-06-30 | 2020-10-09 | 北京鼎泰智源科技有限公司 | Verification code identification method |
| CN112270325A (en) * | 2020-11-09 | 2021-01-26 | 携程旅游网络技术（上海）有限公司 | Character verification code recognition model training method, recognition method, system, device and medium |
| CN112270325B (en) * | 2020-11-09 | 2024-05-24 | 携程旅游网络技术（上海）有限公司 | Character verification code recognition model training method, recognition method, system, equipment and medium |
| CN114817937A (en) * | 2021-01-19 | 2022-07-29 | 北京嘀嘀无限科技发展有限公司 | Keyboard encryption method, device, storage medium and computer program product |
| CN114817893A (en) * | 2021-01-19 | 2022-07-29 | 北京嘀嘀无限科技发展有限公司 | Authentication code image encryption method, device, storage medium and computer program product |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107360137A (en) | Construction method and device for the neural network model of identifying code identification | |
| US11163991B2 (en) | Method and apparatus for detecting body | |
| US10785241B2 (en) | URL attack detection method and apparatus, and electronic device | |
| CN109922032B (en) | Method, device, equipment and storage medium for determining risk of logging in account | |
| CN107679466B (en) | Information output method and device | |
| CN108121984A (en) | A kind of character identifying method and device | |
| CN107204956B (en) | Website identification method and device | |
| CN109005145A (en) | A kind of malice URL detection system and its method extracted based on automated characterization | |
| CN110046622B (en) | Targeted attack sample generation method, device, equipment and storage medium | |
| CN113674140A (en) | Physical countermeasure sample generation method and system | |
| CN110781976B (en) | Extension method of training image, training method and related device | |
| CN107578034A (en) | information generating method and device | |
| CN108416343B (en) | Face image recognition method and device | |
| CN112597495B (en) | Malicious code detection method, system, equipment and storage medium | |
| CN111915618B (en) | Peak response enhancement-based instance segmentation algorithm and computing device | |
| CN104867225A (en) | Banknote face orientation identification method and apparatus | |
| CN111309222B (en) | Sliding block notch positioning and dragging track generation method for sliding block verification code | |
| CN107819790A (en) | The recognition methods of attack message and device | |
| CN105894028A (en) | User identification method and device | |
| CN105956469A (en) | Method and device for identifying file security | |
| CN116311214B (en) | License plate recognition method and device | |
| CN113569852A (en) | Training method and device of semantic segmentation model, electronic equipment and storage medium | |
| CN110705531A (en) | Missing character detection and missing character detection model establishing method and device | |
| CN113919497A (en) | Attack and defense method based on feature manipulation for continuous learning ability system | |
| CN105447566A (en) | Training device and method, and detection 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 | ||
| AD01 | Patent right deemed abandoned |
Effective date of abandoning: 20210312 |
|
| AD01 | Patent right deemed abandoned |