CN113468492A - Verification method and device for verification information and readable storage medium - Google Patents
Verification method and device for verification information and readable storage medium Download PDFInfo
- Publication number
- CN113468492A CN113468492A CN202110790451.XA CN202110790451A CN113468492A CN 113468492 A CN113468492 A CN 113468492A CN 202110790451 A CN202110790451 A CN 202110790451A CN 113468492 A CN113468492 A CN 113468492A
- Authority
- CN
- China
- Prior art keywords
- verification
- verification information
- information
- path
- sequence
- 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
- 238000012795 verification Methods 0.000 title claims abstract description 259
- 238000000034 method Methods 0.000 title claims abstract description 95
- 238000012545 processing Methods 0.000 claims abstract description 12
- 239000011159 matrix material Substances 0.000 abstract description 9
- 230000008569 process Effects 0.000 description 15
- 238000012216 screening Methods 0.000 description 11
- 210000005266 circulating tumour cell Anatomy 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 6
- 238000009826 distribution Methods 0.000 description 6
- 238000012015 optical character recognition Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 238000005259 measurement Methods 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 230000004044 response Effects 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008094 contradictory effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013518 transcription Methods 0.000 description 1
- 230000035897 transcription Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/31—User authentication
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/30—Authentication, i.e. establishing the identity or authorisation of security principals
- G06F21/31—User authentication
- G06F21/32—User authentication using biometric data, e.g. fingerprints, iris scans or voiceprints
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/10—Text processing
- G06F40/12—Use of codes for handling textual entities
- G06F40/126—Character encoding
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Security & Cryptography (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Artificial Intelligence (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention provides a verification method and device for verification information and a readable storage medium. The verification method of the verification information comprises the following steps: receiving verification information sent by a client; processing the verification information to obtain a characteristic sequence of the verification information; and verifying the verification information according to the accuracy and the preset probability of the characteristic sequence. According to the technical scheme provided by the invention, the threshold values of the error recognition rate and the passing rate are set, the original confidence coefficient and inference speed model index relation matrix is changed into three dimensions of risk, passing rate, confidence coefficient and inference speed, the mutual relation between the risk and the confidence coefficient and the inference speed can be balanced without retraining a sequence model, the risk and the passing rate are effectively controlled, the error recognition rate of verification information is reduced, and the accuracy of verification information verification is improved.
Description
Technical Field
The invention relates to the technical field of verification information safety, in particular to a verification method of verification information, a verification device of verification information and a readable storage medium.
Background
In the related art, in a general application scenario, CTC is used for decoding, and the main focus is to improve the confidence of the decoding output of the model and the inference decoding speed. However, in application scenarios such as finance and banking, the two-dimensional measurement indexes are included, and the risk and the throughput rate are also used as the measurement indexes of the third dimension, so that the conventional CTC decoding method cannot balance the interrelation among the risk, the confidence and the inference decoding speed.
Disclosure of Invention
The present invention is directed to solving at least one of the problems of the prior art or the related art.
To this end, a first aspect of the present invention provides a verification method of authentication information.
The second aspect of the invention also provides a verification device for verifying information.
A third aspect of the invention also provides a readable storage medium.
In view of this, a first aspect of the present invention provides a method for verifying verification information, including: receiving verification information sent by a client; processing the verification information to obtain a characteristic sequence of the verification information; and verifying the verification information according to the accuracy and the preset probability of the characteristic sequence.
The verification method of the verification information provided by the invention receives the verification information sent by the user in the process of verifying the legal identity of the user, introduces the received verification information into the sequence model, decodes the verification information in a CTC path search mode to obtain the characteristic sequence of the verification information, and further determines the accuracy of the characteristic sequence in the verification information sent by the user. And setting a preset probability according to a service scene of user identity authentication, wherein the better probability is the passing rate through which the authentication information input by the user can pass, and comparing the accuracy of the characteristic sequence with the preset probability so as to verify the received authentication information. Specifically, in the prior art, the received verification information is decoded through CTC, and only the confidence level of the model decoding output and the metrics of two dimensions of the inference decoding speed of the model are included. However, in the scenes of finance, banking and the like, the method further comprises a measurement index of a third dimension, namely risk and passing rate, if the mutual relation between the risk and the confidence is balanced, the neural network sequence model needs to be trained and deployed again.
It should be understood that the preset probability is a risk and a passing rate threshold set according to an application scenario of user authentication. The risk, namely the misrecognition rate and the pass rate are contradictory, the misrecognition rate and the pass rate are mutually influenced, one index becomes better, the other index becomes worse, and the preset probability is set in a compromise mode according to the attention degree of an actual application scene so as to achieve the purpose of balancing the risk and the pass rate.
Further, in the user identity verification scenario, the legal identity of the user may be verified in the forms of voice Recognition, OCR (Optical Character Recognition), lip language Recognition, and the like, which is not limited herein.
According to the verification method of the verification information provided by the invention, the following additional technical characteristics can be provided:
in the above technical solution, further, the step of processing the verification information to obtain the feature sequence of the verification information specifically includes: inputting the verification information into a sequence model to generate at least one path corresponding to the verification information; comparing the at least one path with a preset path, and determining an effective path matched with the preset path in the at least one path; and determining a characteristic sequence corresponding to the effective path according to the effective path.
In the technical scheme, after verification information sent by a client is received, the verification information is input into a trained sequence model, at least one output path is generated through a path search mode of a CTC (central control unit), the at least one output path is matched with a preset path, an effective path, namely a correct path, in the output path is determined by judging whether all output paths corresponding to the input verification information are matched with the preset path or not, and then a feature sequence corresponding to the effective path is determined to be used as a correct feature sequence, so that the efficiency of verification information judgment is effectively improved.
In any of the above technical solutions, further, before receiving the verification information sent by the client, the method further includes: receiving an authentication request sent by a client, and generating a given character string in response to the authentication request; and sending the given character string to the client so that the client sends the verification information according to the given character string.
In the technical scheme, in a user identity verification scene, after an identity verification request sent by a client is received, a server responds to the identity verification request of a user, a given character string is generated by the server and is used as a verification code or other random information needing the cooperation and discrimination of the user, the generated given character string is sent to the client, after the cooperation and discrimination of the user, verification information is uploaded to the server, then the received verification information is decoded through a sequence model, and then the verification information is verified. The whole process shows that the given character string is known in advance as the information needing screening processing by the legal user, the prior information is fully utilized to verify the verification information matched with the screening by the user, the time for distinguishing the content of the verification information is effectively shortened, and the efficiency of the legal identity verification of the user is improved.
In any of the above technical solutions, further, the verification method for the verification information further includes: inputting a given character string into a sequence model; and generating a preset path according to the given character string.
In the technical scheme, after a given character string is generated in response to an identity verification request of a user, the given character string is input into a sequence model, and for the trained model, a correct path corresponding to the given character string is generated as a preset path according to the prior information of the given character string. The method comprises the steps of inputting a given character string into a sequence model, determining a correct path, and judging an effective path in verification information according to the correct path, so that irrelevant paths generated by the verification information in the model do not need to be judged, and only comparing at least one path generated by the verification information, wherein the path consistent with the correct path is used as the effective path. The original generating form is skillfully changed into the discriminant form, so that the efficiency of verifying information discrimination is improved.
In any of the above technical solutions, further, the step of verifying the verification information according to the accuracy and the preset probability of the feature sequence specifically includes: acquiring the accuracy of the characteristic sequence, and comparing the accuracy with a preset probability; when the accuracy is greater than or equal to the preset probability, the verification information passes the verification; and when the accuracy is smaller than the preset probability, the verification information is not verified.
In the technical scheme, the received verification information is decoded through a CTC (central control unit), an effective path matched with a correct result path of a given character string is determined, a feature sequence corresponding to the effective path is determined as an output result, and the probability that the feature sequence is the correct result as the output result, namely the correct rate of the feature sequence, is obtained. And comparing the accuracy with a preset passing check probability value. If the accuracy is greater than or equal to the preset passing verification probability, that is, the verification information of the user matching with the identity screening is correct, the legal identity verification of the user is passed; if the accuracy is smaller than the preset probability, the verification information of the user matching with the identity screening is wrong, and the legal identity verification of the user is not passed. Through setting a threshold value of the risk balance and the passing rate, verification information of the user matched with identity screening is verified, so that high identification accuracy of user identity verification is achieved.
In any of the above technical solutions, further, the verification method for the verification information further includes: and setting a preset probability according to the application occasion of the verification information.
In the technical scheme, the method comprises the steps of receiving verification information which is sent by a client and used for verifying the legal identity of a user, and setting a preset probability according to a service scene corresponding to the legal identity verification so as to control different attention degrees of different service scenes on safety and passing rate. Compared with the current CTC application scene, the model is mainly deployed based on the confidence coefficient of model decoding output and the reasoning decoding speed of the model, the method changes the original model index relation matrix into three dimensions of the confidence coefficient, the reasoning decoding speed and the risk and passing rate by setting the preset probability, balances the mutual relation between the risk and the confidence coefficient, effectively controls the risk and the passing rate of the legal identity verification of the user, provides system control of a specific application scene, improves service management control and industry supervision, and lays a solid foundation for protecting the identity safety of the user and increasing a safety threshold.
In any of the above solutions, further, the given character string includes at least one of: numbers, english letters, and chinese characters.
In the technical scheme, in a user identity verification scene, the given character string can be any combination of numbers, English letters and Chinese characters, and the flexibility of the legal identity verification of the user is improved by taking the combined given character string as a verification code or random information needing the matching and discrimination of the user.
In any of the above technical solutions, further, the verification information includes at least one of: picture authentication information, voice authentication information, and video authentication information.
In the technical scheme, the legal identity of the user can be verified by picture verification information, voice verification information or video verification information. And performing sequence modeling through the CTC, decoding picture verification information, voice verification information or video verification information, and verifying the content of the verification information so as to verify the legal identity of the user, thereby improving the applicability of the legal identity verification of the user.
According to a second aspect of the present invention, there is provided a verification apparatus for verifying information, comprising: a memory storing a program or instructions; and the processor is connected with the memory and is configured to implement the verification method of the verification information provided by the first aspect when executing the program or the instructions. Therefore, the verification device for verification information has all the beneficial effects of the verification method for verification information provided in the first aspect, and details are not repeated herein.
According to a third aspect of the present invention, there is provided a readable storage medium on which a program or instructions are stored, the program or instructions, when executed by a processor, performing the verification method of the authentication information set forth in the first aspect. Therefore, the readable storage medium has all the advantages of the verification method for the verification information provided in the first aspect, and redundant description is omitted for avoiding repetition.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Drawings
The above and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow chart of a verification method for verification information according to an embodiment of the present invention;
FIG. 2 is a second flowchart illustrating a method for verifying authentication information according to an embodiment of the present invention;
FIG. 3 is a third flowchart illustrating a verification method for verification information according to an embodiment of the present invention;
FIG. 4 is a flow chart of a verification method for verification information according to an embodiment of the present invention;
FIG. 5 is a fifth flowchart illustrating a verification method for verification information according to an embodiment of the present invention;
FIG. 6 shows a sixth flowchart of a verification method for verification information according to an embodiment of the present invention;
FIG. 7 illustrates one of the relational matrix diagrams of the sequence model interest metric of one embodiment of the present invention;
FIG. 8 is a second graph of a relational matrix of the sequence model interest metric according to an embodiment of the present invention;
FIG. 9 illustrates one of the decoding diagrams of authentication information according to an embodiment of the present invention;
FIG. 10 is a second schematic diagram illustrating the decoding of authentication information according to an embodiment of the present invention;
fig. 11 shows a schematic block diagram of a verification device for verifying information according to the present invention.
Wherein, the correspondence between the reference numbers and the names of the components in fig. 11 is:
1100 verification means for verifying the information, 1102 memory, 1104 processor.
Detailed Description
In order that the above objects, features and advantages of the present invention can be more clearly understood, a more particular description of the invention will be rendered by reference to the appended drawings. It should be noted that the embodiments and features of the embodiments of the present application may be combined with each other without conflict.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those specifically described herein, and therefore the scope of the present invention is not limited by the specific embodiments disclosed below.
A verification method of authentication information, a verification apparatus of authentication information, and a readable storage medium according to some embodiments of the present invention are described below with reference to fig. 1 to 11.
Example 1:
as shown in fig. 1, according to an embodiment of the present invention, a verification method for verification information is provided, the method including:
102, receiving verification information sent by a client;
step 104, processing the verification information to obtain a characteristic sequence of the verification information;
and 106, checking the verification information according to the accuracy and the preset probability of the characteristic sequence.
In this embodiment, in the process of verifying the legal identity of the user, the verification information sent by the user is received, the received verification information is imported into the sequence model, the verification information is decoded in a path search manner of a CTC (continuous time series Classification), a feature sequence of the verification information is obtained, and the accuracy of the feature sequence in the verification information sent by the user is further determined. And setting a preset probability according to a service scene of user identity authentication, wherein the better probability is the passing rate through which the authentication information input by the user can pass, and comparing the accuracy of the characteristic sequence with the preset probability so as to verify the received authentication information. Specifically, in the prior art, the received verification information is decoded through CTC, and only the confidence level of the model decoding output and the metrics of two dimensions of the inference decoding speed of the model are included. However, in the scenes of finance, banking and the like, the method further comprises a measurement index of a third dimension, namely risk and passing rate, if the mutual relation between the risk and the confidence is balanced, the neural network sequence model needs to be trained and deployed again.
It should be understood that the preset probability is a risk and a passing rate threshold set according to an application scenario of user authentication. The risk, namely the misrecognition rate and the pass rate are contradictory, the misrecognition rate and the pass rate are mutually influenced, one index becomes better, the other index becomes worse, and the preset probability is set in a compromise mode according to the attention degree of an actual application scene so as to achieve the purpose of balancing the risk and the pass rate.
Further, in the user identity verification scenario, the legal identity of the user may be verified in the forms of voice Recognition, OCR (Optical Character Recognition), lip language Recognition, and the like, which is not limited herein.
Further, the given string includes at least one of: numbers, english letters, and chinese characters. In the user identity verification scene, the given character string can be any combination of numbers, English letters and Chinese characters, and the flexibility of the legal identity verification of the user is improved by taking the combined given character string as a verification code or random information needing the matching and discrimination of the user.
Further, the verification information includes at least one of: picture authentication information, voice authentication information, and video authentication information. The legal identity of the user can be verified by picture verification information, voice verification information or video verification information. And performing sequence modeling through the CTC, decoding picture verification information, voice verification information or video verification information, and verifying the content of the verification information so as to verify the legal identity of the user, thereby improving the applicability of the legal identity verification of the user.
In a specific embodiment, when the user performs the payment operation, the server needs to verify the legal identity of the user. And receiving the voice verification information read by the user, inputting the voice information into the sequence model, decoding the voice verification information, and extracting the voice verification content from the voice verification information. According to the scene of business handling of the user, setting threshold values of the error identification rate and the passing rate of the identity authentication as the preset probability of passing the authentication, determining the accuracy rate of the extracted voice authentication information content, and comparing the accuracy rate with the preset passing probability to verify the legal identity of the user. Only through legal identity authentication, the user can carry out final payment operation, and the safety and the reliability of payment carried out by the user are effectively improved. Further, the given character string may be generated by performing an operation on a pre-stored character library in combination with a time factor according to a pre-stored character algorithm, for example, pre-storing the character library and assigning a unique identification code to each character in the character library, taking the time of receiving the payment request of the user as a time factor, for example, the time of receiving the payment request is 2021 year, 6 month, 7 day, 12 hour, 13 minutes and 36 seconds, recording the number string 20210607121336 as a time factor, dividing the time factor into 5 sections, calling characters with the character identification codes of 2/02, 106/0712 and 1336 in the character library respectively to form the given character string, and reversely reasoning the time of receiving the payment request by the finally generated given character string in combination with the character algorithm.
Example 2:
as shown in fig. 2, according to an embodiment of the present invention, a verification method for verification information is provided, the method including:
and step 210, verifying the verification information according to the accuracy and the preset probability of the characteristic sequence.
In the embodiment, after receiving the verification information sent by the client, the verification information is input into a trained sequence model, at least one output path is generated in a path search mode of a CTC, the at least one output path is matched with a preset path, an effective path, namely a correct path, in the output path is determined by judging whether all the output paths corresponding to the input verification information are matched with the preset path, and then a feature sequence corresponding to the effective path is determined as a correct feature sequence, so that the efficiency of verification information judgment is effectively improved.
In a specific embodiment, as shown in fig. 2 and 3, a character decoding matrix diagram using CTCs is shown, wherein a time series, i.e., a length of a character, is indicated in a horizontal direction. The vertical direction represents the total number of characters used by the algorithm for decoding, namely, the matrix is a collection of all paths which satisfy the target output in a monotonous way from left to right and from top to bottom. By matching all paths with the preset correct path, the effective path in all paths can be determined.
Example 3:
as shown in fig. 3, according to an embodiment of the present invention, a verification method for verification information is provided, the method including:
and 310, checking the verification information according to the accuracy and the preset probability of the characteristic sequence.
In the embodiment, in a user identity verification scene, after an identity verification request sent by a client is received, a server responds to the identity verification request of a user, a given character string is generated by the server and is used as a verification code or other random information needing the cooperation and discrimination of the user, the generated given character string is sent to the client, after the cooperation and discrimination of the user, verification information is uploaded to the server, then the received verification information is decoded through a sequence model, and then the verification information is verified. The whole process shows that the given character string is known in advance as the information needing screening processing by the legal user, the prior information is fully utilized to verify the verification information matched with the screening by the user, the time for distinguishing the content of the verification information is effectively shortened, and the efficiency of the legal identity verification of the user is improved.
In a specific embodiment, the given character string may be a static character or a dynamic character, and after the server issues the given character string to the client, the user may read the content of the given character string to cooperate with the screening, and upload the voice information corresponding to the content of the given character string read by the user to the server, so as to verify the legal identity of the user.
Example 4:
as shown in fig. 4, according to an embodiment of the present invention, a verification method for verification information is provided, which includes:
step 404, sending the given character string to the client so that the client sends verification information according to the given character string;
and step 414, checking the verification information according to the accuracy and the preset probability of the characteristic sequence.
In the embodiment, after a given character string is generated in response to an authentication request of a user, the given character string is input into a sequence model, and for the trained model, a correct path corresponding to the given character string is generated as a preset path according to the prior information of the given character string. The method comprises the steps of inputting a given character string into a sequence model, determining a correct path, and judging an effective path in verification information according to the correct path, so that irrelevant paths generated by the verification information in the model do not need to be judged, and only comparing at least one path generated by the verification information, wherein the path consistent with the correct path is used as the effective path. The original generating form is skillfully changed into the discriminant form, so that the efficiency of verifying information discrimination is improved.
Example 5:
as shown in fig. 5, according to an embodiment of the present invention, a verification method for verification information is provided, which includes:
step 504, processing the verification information to obtain a characteristic sequence of the verification information;
step 506, acquiring the accuracy of the characteristic sequence, and comparing the accuracy with a preset probability;
in step 512, the verification information is not verified.
In this embodiment, the received verification information is decoded through the CTC, an effective path matching with a correct result path of a given character string is determined, a feature sequence corresponding to the effective path is determined as an output result, and a probability that the feature sequence is the correct result as the output result, that is, the correct rate of the feature sequence is obtained. And comparing the accuracy with a preset passing check probability value. If the accuracy is greater than or equal to the preset passing verification probability, that is, the verification information of the user matching with the identity screening is correct, the legal identity verification of the user is passed; if the accuracy is smaller than the preset probability, the verification information of the user matching with the identity screening is wrong, and the legal identity verification of the user is not passed. Through setting a threshold value of the risk balance and the passing rate, verification information of the user matched with identity screening is verified, so that high identification accuracy of user identity verification is achieved.
Example 6:
as shown in fig. 6, according to an embodiment of the present invention, a verification method for verification information is provided, which includes:
step 604, processing the verification information to obtain a characteristic sequence of the verification information;
step 606, verifying the verification information according to the accuracy and the preset probability of the characteristic sequence;
step 608, setting a preset probability according to the application occasion of the verification information.
In the embodiment, the verification information which is sent by the client and used for verifying the legal identity of the user is received, and the preset probability is set according to the service scene corresponding to the legal identity verification so as to control different attention degrees of different service scenes on the safety and the passing rate. Compared with the current CTC application scene, the model is mainly deployed based on the confidence coefficient of model decoding output and the reasoning decoding speed of the model, the method changes the original model index relation matrix into three dimensions of the confidence coefficient, the reasoning decoding speed and the risk and passing rate by setting the preset probability, balances the mutual relation between the risk and the confidence coefficient, effectively controls the risk and the passing rate of the legal identity verification of the user, provides system control of a specific application scene, improves service management control and industry supervision, and lays a solid foundation for protecting the identity safety of the user and increasing a safety threshold.
Example 7:
according to a specific embodiment of the invention, the verification method of the verification information is provided, and the method can be effectively applied to the scenes of banks, finance and the like. By using the method, an ROC (Receiver Operating Characteristic Curve) Curve and a P-R (Precision-Recall Curve) Curve of the model can be obtained, and different preset probabilities can be set in different scenes through setting of proper preset probabilities, so that the purpose of balancing risk and passing rate can be achieved without retraining the network model.
With the development of information technology, man-machine interaction presents more and more forms. With the development of AI (Artificial Intelligence) technology with deep learning as a core, some human-computer interaction modalities which were user-friendly but limited in algorithmic performance have been possible. Among them, the technology is widely used in speech recognition, OCR recognition, lip language recognition, etc. The common characteristic of the method is a sequence learning model based on a generating formula. Data alignment is a very central problem in sequence learning, and the current mainstream alignment method is to use CTCs.
Specifically, CTCs can solve the problem of training data requiring alignment in a dynamic programming-like manner. The alignment problem is how the input Label (Label) corresponds to a Feature (Feature) learned by a neural network or other modeling method. Taking speech recognition as an example, before model training, we have an audio clip and its corresponding transcription text, i.e., Label. How the characters in the Label correspond to the features (features) of the audio is a problem to be considered in the alignment method. The above discussion may be formalized as a mapping:
f: x → Y, wherein X ∈ RT,Y∈RU(1),
In the above mapping relationship, on one hand, the dimensions T and U of X and Y may be varied and are not necessarily the same; on the other hand, it is also difficult to obtain an accurate correspondence between X and Y in actual data. While the sequence modeling problem can be abstracted to compute the conditional probability P (Y | X).
Further, CTC solves the above-mentioned problem that dimensions T and U of X and Y may vary and not necessarily be the same by introducing a null character (Blank). And the introduction of the empty character can also effectively solve the problem of repeated character output. And the problem that the accurate corresponding relation between X and Y in actual data is difficult to obtain is solved through a dynamic planning or an improved method based on directional Search (Beam Search), namely the probability and summation (Sum) of feasible paths between input X and output Y can be approximately estimated.
Specifically, one brief process of CTC computation alignment, exemplified by speech recognition, is as follows: a sequence of speech spectra, which may be an original audio clip (clip) cut at regular intervals, is input and then MFCCs (Mel-frequency cepstral coefficients ) are solved or extracted using a neural network.
Further, the input is fed into the RNN (Recurrent Neural Network), and for each minimum time scale, i.e. time step (Timestep), modeled by the Network, the RNN Network can give a conditional probability distribution p at each time stepl(a | X), where a is a feasible character in Y, including a null character (blank), and l is the corresponding time step.
Further, after the Output Distribution (Output Distribution) at each time step is obtained, the probability of different sequence combinations can be calculated.
Further, the final output probability distribution is obtained by calculating different alignments (alignments), that is, edge probability distributions of possible combinations of output paths. The whole process can be expressed by the following formula:
further, the conventional usage method of CTCs is mainly used as a Loss Function (Loss Function) of a neural network or a Hidden Markov Model (HMM). Specifically, after the model is trained, the model itself is decoded by a path Search method of CTC using the Generative (Generative) property of the model, or the decoded output TopN (the maximum N records in each group) is controlled by combining with a Beam Search (Beam Search) technique. The overall process is generative. This approach is relatively natural and therefore very widely used, by learning the intrinsic distribution of the data and then combining the CTCs for decoding output. However, the main focus of the general application scenario is to improve the single-dimensional metric such as Confidence (Confidence) of the model decoding output, or sometimes include the metric such as Inference (Inference) decoding speed of the model. However, some application scenarios include not only the two-dimensional metrics but also a third-dimensional metric, risk and throughput, which are particularly interesting in finance, banking, and other scenarios, and simply, the metric may balance the relationship between risk and confidence by setting a Threshold (Threshold).
The CTC decoding result verification method based on the discriminant provided by the invention is an improvement for balancing the passing rate and the risk in the standard CTC decoding process in consideration of the verification scenes of finance, banks and the like. As shown in fig. 7 and 8, in the current scenario of using CTCs, model training and deployment are mainly performed based on an index relationship matrix of confidence and inference speed, but the method provided by the present invention can change the original model index relationship matrix into three dimensions, and can effectively control risk and throughput rate by adjusting a threshold.
Specifically, in a verification scene, the server issues some passwords, such as verification codes or other random information which needs to be matched and discriminated by the user, and after the user is matched and discriminated, the information is uploaded to the server and then processed through a neural network model. The information that legal users need to be screened is known in advance, and the traditional verification method only simply re-identifies the information uploaded by the users and does not fully utilize the important prior information. The prior information is fully utilized, the original generating scheme can be ingeniously changed into a discriminant scheme, and in addition, different attention degrees of different scenes on safety and passing rate can be controlled through preset probability.
As shown in fig. 9 and fig. 10, the process of path search in the standard CTC decoding process is described, wherein the CTC decoding process utilizes a dynamic programming-like method, each path has a confidence score, and since the priori of the CTC modeling is the Independence (Independence) of the probability outputs of each time step, the confidence score of a path may be the Likelihood Product (Likelihood Product) of each time step of the path, i.e., the concatenated Product part in equation (2). Further, the CTC introduces the concept of null characters, i.e., one more null character in the decoded word stock, to solve the alignment problem and the problem of consecutive repeated character output. The training process of CTC is a problem of continuously finding the optimal path, and since the introduction of null characters causes many identical output paths, the final output is the sum of these paths, i.e. the consecutive sign in formula (2).
Further, in the verification scenario, for the trained model, since a correct result is known, it is not necessary to actually determine an irrelevant path, that is, only to check whether the path likelihood is consistent with the verification information issued by the server. As shown in fig. 10, the deepened solid line is a path consistent with the issued verification information, and there may be multiple paths, but if the word bank is determined in advance, the confidence probability can be easily obtained through the search rule of the CTC, that is, the verification problem finally becomes the form of the following discriminant:
the result form is similar to the output of a classification problem, the ROC or P-R curve of the model can be calculated by using the method, and then the aim of balancing the risk and the passing rate can be achieved by controlling the threshold value in combination with a specific business scene.
Furthermore, the method combines prior information in an actual application scene, namely the correct result of model reasoning is known, converts the original CTC generation decoding problem into a discriminant problem, and further can achieve the problem of balancing risk and passing rate through the preset probability, namely the setting of a threshold value. The method has important application value for the scene of identity verification by using the sequence type model.
In particular, the scheme can be used for the problem of judging the identity of the user by utilizing voice, OCR or lip language in a verification scene.
In addition, in some current scenes, sequence modeling can be performed without using CTCs, such as a motion discrimination problem in face recognition or a dazzling living body problem, multi-frame image information can be simply input, motion information of a user is extracted by using 3D space convolution, and then result judgment is performed by using a classification network. However, the reliability of this scheme is very dependent on the SDK (Software Development Kit) of the algorithm, and the output result is significantly affected by the class increase. The flexibility is inferior to the solution proposed by the present invention, which can be used not only on SDK but also on H5 (hypertext markup language revision 5) scenario.
Example 8:
as shown in fig. 11, according to an embodiment of the second aspect of the present invention, there is provided a verification apparatus 1100 for verifying information, including: a memory 1102, the memory 1102 storing programs or instructions; the processor 1104 is connected to the memory 1102, and the processor 1104 is configured to implement the verification method of the verification information proposed in the first aspect when executing the program or the instructions. Therefore, the verification device for verification information has all the beneficial effects of the verification method for verification information provided in the first aspect, and details are not repeated herein.
Example 11:
according to a third aspect of the present invention, there is provided a readable storage medium on which a program or instructions are stored, the program or instructions, when executed by a processor, performing the verification method of the authentication information set forth in the first aspect. Therefore, the readable storage medium has all the advantages of the verification method for the verification information provided in the first aspect, and redundant description is omitted for avoiding repetition.
In the present invention, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element. Further, it should be noted that the scope of the methods and apparatus of the embodiments of the present application is not limited to performing the functions in the order illustrated or discussed, but may include performing the functions in a substantially simultaneous manner or in a reverse order based on the functions involved, e.g., the methods described may be performed in an order different than that described, and various steps may be added, omitted, or combined. In addition, features described with reference to certain examples may be combined in other examples.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present application may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the methods of the embodiments of the present application.
In the description herein, the description of the terms "one embodiment," "some embodiments," "specific embodiments," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A verification method for verification information is characterized by comprising the following steps:
receiving verification information sent by a client;
processing the verification information to obtain a characteristic sequence of the verification information;
and checking the verification information according to the accuracy and the preset probability of the characteristic sequence.
2. The method for verifying the verification information according to claim 1, wherein the step of processing the verification information to obtain the feature sequence of the verification information specifically includes:
inputting the verification information into a sequence model to generate at least one path corresponding to the verification information;
comparing at least one path with a preset path, and determining an effective path matched with the preset path in the at least one path;
and determining the characteristic sequence corresponding to the effective path according to the effective path.
3. The method for verifying the verification information according to claim 1, wherein before receiving the verification information sent by the client, the method further comprises:
receiving an authentication request sent by the client, and responding to the authentication request to generate a given character string;
and sending the given character string to the client so that the client sends the verification information according to the given character string.
4. The verification method of authentication information according to claim 3, further comprising:
inputting the given character string into a sequence model;
and generating a preset path according to the given character string.
5. The method for verifying the verification information according to claim 1, wherein the step of verifying the verification information according to the correctness and the preset probability of the feature sequence specifically comprises:
acquiring the accuracy of the characteristic sequence, and comparing the accuracy with the preset probability;
when the accuracy is larger than or equal to the preset probability, the verification information passes the verification;
and when the accuracy is smaller than the preset probability, the verification information is not verified.
6. The verification method of authentication information according to any one of claims 1 to 5, further comprising:
and setting the preset probability according to the application occasion of the verification information.
7. The verification method of authentication information according to any one of claims 1 to 5,
the given string includes at least one of: numbers, english letters, and chinese characters.
8. The verification method of authentication information according to any one of claims 1 to 5,
the authentication information includes at least one of: picture authentication information, voice authentication information, and video authentication information.
9. A verification apparatus for verifying information, comprising:
a memory storing a program or instructions;
a processor coupled to the memory, the processor implementing the method of verifying authentication information of any of claims 1 to 8 when executing the program or instructions.
10. A readable storage medium on which a program or instructions are stored, characterized in that said program or instructions, when executed by a processor, implement the steps of the verification method of verification information according to any one of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110790451.XA CN113468492A (en) | 2021-07-13 | 2021-07-13 | Verification method and device for verification information and readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110790451.XA CN113468492A (en) | 2021-07-13 | 2021-07-13 | Verification method and device for verification information and readable storage medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN113468492A true CN113468492A (en) | 2021-10-01 |
Family
ID=77880244
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110790451.XA Pending CN113468492A (en) | 2021-07-13 | 2021-07-13 | Verification method and device for verification information and readable storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113468492A (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104836777A (en) * | 2014-02-10 | 2015-08-12 | 腾讯科技(深圳)有限公司 | Identity verification method and system |
CN105024816A (en) * | 2014-04-25 | 2015-11-04 | 腾讯科技(深圳)有限公司 | Information verification method based on audio frequency and device |
US20160127219A1 (en) * | 2014-11-05 | 2016-05-05 | Huawei Technologies Co., Ltd. | System and Method for Data Path Validation and Verification |
US20170255764A1 (en) * | 2014-11-21 | 2017-09-07 | Alibaba Group Holding Limited | Identity information verification |
CN110852427A (en) * | 2020-01-15 | 2020-02-28 | 支付宝(杭州)信息技术有限公司 | Evidence obtaining environment verification method and device and electronic equipment |
CN111340117A (en) * | 2020-02-27 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | CTC model training method, data processing method, device and storage medium |
CN111669350A (en) * | 2019-03-05 | 2020-09-15 | 阿里巴巴集团控股有限公司 | Identity verification method, verification information generation method, payment method and payment device |
CN112651010A (en) * | 2020-12-28 | 2021-04-13 | 深圳赛安特技术服务有限公司 | Method and device for verifying sliding verification code, computer equipment and medium |
-
2021
- 2021-07-13 CN CN202110790451.XA patent/CN113468492A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104836777A (en) * | 2014-02-10 | 2015-08-12 | 腾讯科技(深圳)有限公司 | Identity verification method and system |
CN105024816A (en) * | 2014-04-25 | 2015-11-04 | 腾讯科技(深圳)有限公司 | Information verification method based on audio frequency and device |
US20160127219A1 (en) * | 2014-11-05 | 2016-05-05 | Huawei Technologies Co., Ltd. | System and Method for Data Path Validation and Verification |
US20170255764A1 (en) * | 2014-11-21 | 2017-09-07 | Alibaba Group Holding Limited | Identity information verification |
CN111669350A (en) * | 2019-03-05 | 2020-09-15 | 阿里巴巴集团控股有限公司 | Identity verification method, verification information generation method, payment method and payment device |
CN110852427A (en) * | 2020-01-15 | 2020-02-28 | 支付宝(杭州)信息技术有限公司 | Evidence obtaining environment verification method and device and electronic equipment |
CN111340117A (en) * | 2020-02-27 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | CTC model training method, data processing method, device and storage medium |
CN112651010A (en) * | 2020-12-28 | 2021-04-13 | 深圳赛安特技术服务有限公司 | Method and device for verifying sliding verification code, computer equipment and medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11210470B2 (en) | Automatic text segmentation based on relevant context | |
Gomez-Alanis et al. | On joint optimization of automatic speaker verification and anti-spoofing in the embedding space | |
CN110147445A (en) | Intension recognizing method, device, equipment and storage medium based on text classification | |
CN113204952B (en) | Multi-intention and semantic slot joint identification method based on cluster pre-analysis | |
Davel et al. | Pronunciation prediction with default&refine | |
CN112992126B (en) | Voice authenticity verification method and device, electronic equipment and readable storage medium | |
CN112016553B (en) | Optical Character Recognition (OCR) system, automatic OCR correction system, method | |
CN104167208A (en) | Speaker recognition method and device | |
CN112800239A (en) | Intention recognition model training method, intention recognition method and device | |
US11790018B1 (en) | Apparatus for attribute traversal | |
Ma et al. | Spoken language recognition using ensemble classifiers | |
Nazir et al. | A computer-aided speech analytics approach for pronunciation feedback using deep feature clustering | |
Singh et al. | Correlation‐based classifier combination in the field of pattern recognition | |
CN114530145A (en) | Speech recognition result error correction method and device, and computer readable storage medium | |
CN114586038A (en) | Method, device, equipment and medium for event extraction and extraction model training | |
CN111694936A (en) | Method and device for identifying AI intelligent interview, computer equipment and storage medium | |
US20240028952A1 (en) | Apparatus for attribute path generation | |
CN114548104B (en) | Feature and category intervention-based few-sample entity identification method and model | |
Sheng et al. | Semantic-preserving abstractive text summarization with siamese generative adversarial net | |
CN116362242A (en) | Small sample slot value extraction method, device, equipment and storage medium | |
CN113723111B (en) | Small sample intention recognition method, device, equipment and storage medium | |
CN112133291A (en) | Language identification model training, language identification method and related device | |
KR102363958B1 (en) | Method, apparatus and program for analyzing customer perception based on double clustering | |
CN113468492A (en) | Verification method and device for verification information and readable storage medium | |
KR102443629B1 (en) | Solution and system for news positive tendency analysis using deep learning nlp model |
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 |