CN108647874A - Threshold value determines method and device - Google Patents
Threshold value determines method and device Download PDFInfo
- Publication number
- CN108647874A CN108647874A CN201810421678.5A CN201810421678A CN108647874A CN 108647874 A CN108647874 A CN 108647874A CN 201810421678 A CN201810421678 A CN 201810421678A CN 108647874 A CN108647874 A CN 108647874A
- Authority
- CN
- China
- Prior art keywords
- threshold value
- curve
- appraisal curve
- point
- value
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Educational Administration (AREA)
- Operations Research (AREA)
- Marketing (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
A kind of threshold value of offer of the embodiment of the present invention determines method and device, belongs to verification identification field.This method includes：Obtain the first threshold value, first threshold value is based on determined by the second threshold value and threshold value appraisal curve, threshold value appraisal curve is generated based on the evaluation index for influencing scene verification result under current scene, and the second threshold value is based on determined by sigmoid functions and threshold value appraisal curve；First threshold value and the second threshold value are weighted averagely, using summed result as final threshold value.Due to can be combined sigmoid functions and threshold value appraisal curve during determination final threshold value, to may make output result unique.At the same time, average by the way that the first threshold value and the second threshold value to be weighted, the accuracy and robustness of final threshold value can be improved.
Description
Technical field
The present embodiments relate to verification identification fields, and method and device is determined more particularly, to a kind of threshold value.
Background technology
Biometric technology is by more and more system applications at present, either in punch card system, shopping payment system
A threshold value is typically arranged in system or brush face authentication system, can when system is when getting data to be verified
Calculate the corresponding score of data to be verified.Determine whether the data to be verified can be verified with threshold value based on the score
Pass through.
In the related art, typically threshold value evaluation is generated according to the evaluation index of verification result is influenced in system
Curve, such as：ROC curve, DET curves and PR curves.It is directly true by sigmoid functions that it is based on threshold value appraisal curve again
Determine only one threshold value, determining threshold value accuracy and robustness are poor.
Invention content
To solve the above-mentioned problems, the embodiment of the present invention provides one kind and overcoming the above problem or solve at least partly
The threshold value for stating problem determines method and device.
According to a first aspect of the embodiments of the present invention, a kind of threshold value is provided and determines that method, this method include：
The first threshold value is obtained, the first threshold value is true based on the second threshold value and threshold value appraisal curve institute
Fixed, threshold value appraisal curve is generated based on the evaluation index for influencing scene verification result under current scene, second
Threshold value is based on determined by sigmoid functions and threshold value appraisal curve；
Based on the corresponding weighted value of the first threshold value and the corresponding weighted value of the second threshold value, to the first thresholding threshold
Value and the second threshold value are weighted averagely, using summed result as final threshold value.
Method provided in an embodiment of the present invention, by the first threshold value of acquisition and the second threshold value, then by first
Limit threshold value and the second threshold value are weighted averagely, with the threshold value that determination is final.Wherein, the first threshold value is to be based on
Determined by second threshold value and threshold value appraisal curve, and the second threshold value is to be based on sigmoid functions and door
It limits determined by threshold ratings curve.Due to can be by sigmoid functions and thresholding threshold during determination final threshold value
Value appraisal curve is combined, to may make output result unique.At the same time, by by the first threshold value and second
Limit threshold value is weighted averagely, and the accuracy and robustness of final threshold value can be improved.
Second aspect according to embodiments of the present invention, provides a kind of threshold value determining device, which includes：
Acquisition module, for obtaining the first threshold value, the first threshold value is to be based on the second threshold value and thresholding threshold
It is worth determined by appraisal curve, threshold value appraisal curve is based on the evaluation index for influencing scene verification result under current scene
It is generated, the second threshold value is based on determined by sigmoid functions and threshold value appraisal curve；
Determining module, for being based on the corresponding weighted value of the first threshold value and the corresponding weight of the second threshold value
Value is weighted averagely, using summed result as final threshold value the first threshold value and the second threshold value.
According to a third aspect of the embodiments of the present invention, a kind of scene verification method is provided, including：
The corresponding score of data to be verified is obtained, score is that data to be verified are input to the corresponding verification system of current scene
It is obtained after system；
If score is more than threshold value, it is determined that for data to be verified by verification, threshold value is by above-mentioned threshold value
It determines determined by method.
According to a fourth aspect of the embodiments of the present invention, it provides a kind of threshold value and determines equipment, including：
At least one processor；And
At least one processor being connect with processor communication, wherein：
Memory is stored with the program instruction that can be executed by processor, and the instruction of processor caller is able to carry out first party
The threshold value that any possible realization method is provided in the various possible realization methods in face determines method.
According to a fifth aspect of the embodiments of the present invention, a kind of non-transient computer readable storage medium is provided, it is non-transient
Computer-readable recording medium storage computer instruction, computer instruction make the various possible realities of computer execution first aspect
The threshold value that any possible realization method is provided in existing mode determines method.
It should be understood that above general description and following detailed description is exemplary and explanatory, it can not
Limit the embodiment of the present invention.
Description of the drawings
Fig. 1 is that a kind of threshold value of the embodiment of the present invention determines the flow diagram of method；
Fig. 2 is the method flow schematic diagram that a kind of scene of the embodiment of the present invention is verified；
Fig. 3 is a kind of structural schematic diagram of threshold value determining device of the embodiment of the present invention；
Fig. 4 is that a kind of threshold value of the embodiment of the present invention determines the structural schematic diagram of equipment.
Specific implementation mode
With reference to the accompanying drawings and examples, the specific implementation mode of the embodiment of the present invention is described in further detail.With
Lower embodiment is not limited to the range of the embodiment of the present invention for illustrating the embodiment of the present invention.
In the related art, threshold value is determined usually using two ways.A kind of mode is to be made according to system
Scene obtains the positive counter-example contextual data that precise requirements are adapted quantity according to the precise requirements of data to be verified.
And positive counter-example contextual data is inputted into verification system, corresponding evaluation index under each score threshold is exported, is referred to further according to evaluation
Plotting threshold ratings curve, and using the position of equalization point in threshold ratings curve as final threshold value.Another kind side
Formula is that the threshold ratings curve drawn evaluation index using sigmoid functions is regular on smooth threshold curve, then by
The preconfigured parameter value of sigmoid functions, exports final threshold value.
But in above-mentioned existing two schemes, the first scheme directly uses the position of equalization point in threshold ratings curve to make
For final threshold value, when the position of equalization point in threshold ratings curve has multiple, which equalization point is system can not judge
Can be relatively low as final threshold value, or the final threshold value accuracy judged, do not meet actual requirement.Second
Scheme directly exports a final threshold value using sigmoid functions, and robustness is poor during use in system, and
It is required for again manual test, calculating and adjustment, the processing procedure of system cumbersome for each different scene.
For said circumstances, a kind of threshold value of offer of the embodiment of the present invention determines method.Referring to Fig. 1, this method includes：
101, the first threshold value is obtained, the first threshold value is to be based on the second threshold value and threshold value appraisal curve
Identified, threshold value appraisal curve is generated based on the evaluation index for influencing scene verification result under current scene,
Second threshold value is based on determined by sigmoid functions and threshold value appraisal curve.
In a step 101, threshold value appraisal curve is referred to based on the evaluation for influencing scene verification result under current scene
What mark was generated, the evaluation index for influencing scene verification result can be misclassification rate and recall rate.Wherein, each threshold value is equal
Corresponding one group of misclassification rate and recall rate, and can be respectively as abscissa and ordinate.The corresponding misclassification rate of multiple threshold values and
Recall rate can draw out threshold value appraisal curve.The corresponding misclassification rate of each threshold value and recall rate can be by a large amount of
Data statistics obtains.For any threshold value, misclassification rate indicate under the conditions of the threshold value should not by verify data
But the statistic frequency passed through, is represented by with formula：Wherein, FAR is misclassification rate, and NFA is to test
Should not be by verification but by the number of verification when card, NIRA is verification total degree.Recall rate indicates the threshold value item
Under part by accounting of the verification number in verifying total degree.
It is understood that the misclassification rate requirement of verification system is higher, then then proving that the safety of the verification system is wanted
Ask higher.For example, the verification system for amusement that safety is not high, may only require centesimal misclassification rate, and it is used for
The verification system of the applications such as protection, finance may just need the misclassification rate for reaching one thousandth or a ten thousandth.Certainly, it removes
Except using misclassification rate and recall rate as evaluation index, other evaluation indexes, such as accuracy and error rate can also be used,
The embodiment of the present invention is not especially limited this.For ease of description, the present invention is using misclassification rate and recall rate as evaluation index
For illustrate.
Second threshold value is based on determined by sigmoid functions and threshold value appraisal curve.Specifically, first sharp
It is with sigmoid functions that threshold value appraisal curve is regular in smoothing interval, it is defeated further according to the preset parameter values of sigmoid
Go out unique thresholding and as the second threshold value.Wherein, smooth section can value [0,1], the canonical function of sigmoid functions
For：In view of whole threshold value appraisal curve might not can be regular in smoothing interval, the present invention
Embodiment on the basis of canonical function by adding a, and two parameters of b are translated and stretched, after being converted
Sigmoid functions：To choose two points according to demand in curve again：(y1, y1′)、(y2, y2'), and
The two points are substituted intoThe value of a and the value of b is calculated.Finally, taking for y ' is adjusted according to actual demand
Value, to calculate the value of y, and using the value of y as the second threshold value.Since regular smoothing interval is [0,1], so it is excellent
The value of y ' is set as 0.6 by selection of land, and expression reaches 60 points i.e. it is believed that verify data passes through.Certainly, the specific values of y ' can
It is adjusted according to actual conditions, the embodiment of the present invention is not especially limited this.
102, it is based on the corresponding weighted value of the first threshold value and the corresponding weighted value of the second threshold value, to first
Limit threshold value and the second threshold value are weighted averagely, using summed result as final threshold value.
In a step 102, the corresponding weighted value of the first threshold value and the corresponding weighted value of the second threshold value can roots
It is configured according to demand, the embodiment of the present invention is not especially limited this.It wherein, can be by when the first threshold value is even more important
The weight accounting of one threshold value distribution increases, and can distribute the second threshold value when the evaluation of the second threshold value is even more important
Weight accounting increase, the corresponding weighted value of the first threshold value weighted value corresponding with the second threshold value and be 1.Specifically
Ground changes at some in less conventional scenario, such as fingerprint recognition on and off duty and recognition of face scene daily, due to scene compared with
Less to fix and changing, the result that the second threshold value calculates, which has been able to compare, accurately reflects verification result, so as to
The weight accounting of second threshold value is set as 0.8 or 0.9, it might even be possible to be directly disposed as 1.And it is more multiple in some variations
When miscellaneous scene, such as the iris recognition scene of bank custody, since scene is using less, and security performance is more demanding, single
The result calculated using the second threshold value cannot function as final threshold value, thus then can be according to actual conditions by first
The weight accounting for limiting threshold value is turned up, such as the weight accounting of the first threshold value is set as 0.6 or 0.7 etc..In addition, some fields
It, also can be directly by the weight accounting of the first threshold value and the when scape can not carry out weight accounting reasonable distribution according to actual conditions
The weight accounting of two threshold values is both configured to 0.5, and the embodiment of the present invention is not especially limited this.
Method provided in an embodiment of the present invention, by the first threshold value of acquisition and the second threshold value, then by first
Limit threshold value and the second threshold value are weighted averagely, with the threshold value that determination is final.Wherein, the first threshold value is to be based on
Determined by second threshold value and threshold value appraisal curve, and the second threshold value is to be based on sigmoid functions and door
It limits determined by threshold ratings curve.Due to can be by sigmoid functions and thresholding threshold during determination final threshold value
Value appraisal curve is combined, to may make output result unique.At the same time, by by the first threshold value and second
Limit threshold value is weighted averagely, and the accuracy and robustness of final threshold value can be improved.
By the content of above-described embodiment it is found that the embodiment of the present invention can be based on sigmoid functions and threshold value is commented
Valence curve determines the second threshold value, and can determine the first thresholding based on the second threshold value and threshold value appraisal curve
Threshold value.Content based on above-described embodiment, as a kind of alternative embodiment, the embodiment of the present invention is not to obtaining the first threshold value
Mode specifically limit, including but not limited to：Based on threshold value point and several equalization points on threshold value appraisal curve
Distance relation, determine the first threshold value, threshold value point is that the second threshold value corresponds on threshold value appraisal curve
Point, equalization point is the point of contact between the tangent line and threshold value appraisal curve of default slope.
Wherein, threshold value point is that the second threshold value corresponds to the point on threshold value appraisal curve.By above-mentioned implementation
Example content is it is found that the embodiment of the present invention can export the second threshold value by sigmoid functions on threshold ratings curve, then and the
The coordinate that two threshold values correspond on threshold value appraisal curve is threshold value point.Equalization point is the tangent line of default slope
With the point of contact between threshold value appraisal curve, the tangent line for presetting slope can be corresponding same in threshold value appraisal curve
Under coordinate system, several parallel lines of drafting.If there are the tangent lines of any one default slope and threshold value appraisal curve
Between there are point of contacts, then can be using the point of contact as equalization point.
Method provided in an embodiment of the present invention evaluates song with several equalization points by being based on threshold value point in threshold value
Distance relation on line determines the first threshold value.Due to equalization point might have it is multiple, due to be based on threshold value point with it is flat
Distance relation of the weighing apparatus point on threshold value appraisal curve, it may be determined that closest in the value of threshold value with threshold value point
Point so that determine the first threshold value it is more accurate.
Content based on above-described embodiment, as a kind of alternative embodiment, the embodiment of the present invention is not to being based on threshold value
Distance relation of the point with several equalization points on threshold value appraisal curve, determines that the mode of the first threshold value specifically limits
It is fixed, including but not limited to：Threshold value point is calculated at a distance from each equalization point is on threshold value appraisal curve；By most short distance
From corresponding equalization point as target equilibrium point, target equilibrium point is that the first threshold value corresponds on threshold value appraisal curve
Point.
Wherein, point of the first threshold value of target equilibrium point expression correspondence on threshold value appraisal curve, and thresholding threshold
There is correspondence in the dotted or gate limit threshold value on value appraisal curve, to can be obtained the first thresholding threshold based on target equilibrium point
Value.It is understood that equalization point can reflect that each equalization point is approximate with threshold value point at a distance from threshold value point
Degree, the closer equalization point of distance prove that the diversity factor between the equalization point and threshold value point is smaller.And the embodiment of the present invention
It, can also be based on apart from nearest balance under the premise of ensureing close to true verification result by selecting the closer equalization point of distance
Point is modified thresholding threshold point.So that the first threshold value determined is more accurate.
Method provided in an embodiment of the present invention, by regarding the corresponding equalization point of the shortest distance as target equilibrium point, namely
Based on being modified to thresholding threshold point apart from nearest equalization point, so that the first threshold value determined is more accurate.
In view of threshold value appraisal curve has multiple types, each threshold value appraisal curve is according to evaluation index
Difference, the threshold value appraisal curve drawn out also differ.In addition, each threshold value appraisal curve is balanced a determination
When the tangent slope that uses be all different.The tangent slope used when for example, determining equalization point using PR curves is -1.Based on upper
The content for stating embodiment, as a kind of alternative embodiment, it is 1 to preset slope；Correspondingly, threshold value appraisal curve is ROC bent
Line.
Wherein, the tangent line and threshold value appraisal curve that it is 1 that an embodiment of the present invention provides a kind of based on default slope are true
Determine the mode of equalization point.In ROC curve provided in an embodiment of the present invention, the physical meaning that abscissa represents is misclassification rate, is indulged
The physical meaning that coordinate represents is recall rate, and the physical meaning of the equalization point on ROC curve for the misclassification rate at equalization point and is called together
The rate of returning can reach active balance, and the amplitude of variation of misclassification rate and the amplitude of variation of recall rate are consistent substantially.Correspondingly, to make
The amplitude of variation of the amplitude of variation and recall rate that obtain misclassification rate is consistent, then cutting where representing the point on ROC curve
Line slope need to be equal to one.
Method provided in an embodiment of the present invention, the point of contact of the tangent line and ROC curve that are 1 by using slope is as balance
Point ensure that the amplitude of variation of misclassification rate and the amplitude of variation of recall rate are consistent at equalization point, at equalization point
Misclassification rate and recall rate can reach active balance.
In view of the first threshold value and the second threshold value are related with different scenes when distributing weighted value, and scene is special
Sign have repeatability, if every time verification when be required for according to concrete scene determine the corresponding weighted value of the first threshold value and
The corresponding weighted value of second threshold value will then waste a large amount of processing time, and may not necessarily obtain best weighted value.It is based on
The content of above-described embodiment is based on the corresponding weighted value of the first threshold value and the second thresholding as a kind of alternative embodiment
The corresponding weighted value of threshold value is weighted averagely the first threshold value and the second threshold value, using summed result as final
Threshold value before, further include：By the curvilinear characteristic of threshold value appraisal curve, the first threshold value and the second threshold value
It is input to curve discrimination model, determines the corresponding weighted value of the first threshold value and the corresponding weighted value of the second threshold value.
Wherein, the curvilinear characteristic of threshold value appraisal curve is the feature for reflecting the plots changes and amplitude of variation,
Preferably, the curvilinear characteristic of every threshold value appraisal curve can be carried out with the form of the mathematic(al) representation comprising several parameters
It indicates.Curve discrimination model is that the embodiment of the present invention trains in advance, and CNN specifically can be used in the model for determining weighted value
Neural network, RNN neural networks or SVM classifier, the embodiment of the present invention are not especially limited this.It needs to illustrate
Be, the type of curve discrimination model can with disaggregated model, can also regression model, the embodiment of the present invention do not limit this specifically
It is fixed.When curve discrimination model is disaggregated model, by the curvilinear characteristic of the corresponding threshold value appraisal curve of current scene, first
Threshold value and the second threshold value are input to curve discrimination model, the exportable sample to match with threshold value appraisal curve
The corresponding classification of threshold value appraisal curve.Wherein, the sample door of plurality of classes can be pre-set before executing the above process
Threshold ratings curve is limited, the sample threshold value appraisal curve corresponding to the classification of curve discrimination model output is that curve differentiates mould
The immediate curve of threshold value appraisal curve corresponding with current scene in type.
After obtaining the corresponding classification of sample threshold value appraisal curve, the sample threshold value appraisal curve pair can be based on
The classification answered determines the corresponding weighted value of the first threshold value and the corresponding weighted value of the second threshold value.It specifically, can be pre-
It is first sample threshold value appraisal curve one group of weighted value of setting of each classification.Curve discrimination model is in output sample thresholding threshold
It, can be corresponding respectively as the first threshold value by the corresponding one group of weighted value of the category after being worth the corresponding classification of appraisal curve
Weighted value and the corresponding weighted value of the second threshold value.Wherein, the sample threshold value appraisal curve of each classification is corresponding
One group of weighted value includes two weighted values, and two weighted values are corresponding between the first threshold value and the second threshold value
Relationship predefine, and two weighted values and be 1.
It is when curve discrimination model is regression model, the curve of the corresponding threshold value appraisal curve of current scene is special
Sign, the first threshold value and the second threshold value are input to curve discrimination model, the corresponding weight of exportable first threshold value
Value or the corresponding weighted value of the second threshold value.Since the corresponding weighted value of the first threshold value is corresponding with the second threshold value
Weighted value and be 1, to either export the corresponding weighted value of the first threshold value, or the second threshold value of output corresponds to
Weighted value, the finally available corresponding weighted value of first threshold value or the corresponding weighted value of the second threshold value.
Method provided in an embodiment of the present invention, by by the curvilinear characteristic of threshold value appraisal curve, the first threshold value
And second threshold value be input to curve discrimination model, determine the corresponding weighted value of the first threshold value and the second threshold value
Corresponding weighted value, so as to quickly and accurately determine weighted value.
By the content of above-described embodiment it is found that an embodiment of the present invention provides a kind of curve discrimination model, can will input
Threshold value appraisal curve curvilinear characteristic, the first threshold value and the second threshold value, be directly output as and threshold value
The sample threshold value appraisal curve that appraisal curve matches, so as to advance training curve discrimination model.Based on above-mentioned implementation
The content of example, as a kind of alternative embodiment, curve discrimination model is by the sample thresholding threshold under different types of historic scenery
Value appraisal curve is input to disaggregated model or regression model be trained after obtain.
Specifically, the historical data of a large amount of different type scenes can be obtained in advance, and the historical data is with threshold value
The form of appraisal curve is stored, and using these historical datas of storage as sample threshold value appraisal curve.Further
Ground, according to the different scenes feature corresponding to these sample threshold value appraisal curves, by sample threshold value appraisal curve into
Row classification.Wherein, a kind of scene being likely to occur is corresponded to per a kind of sample threshold value appraisal curve, and per a kind of sample thresholding
The first threshold value and the second threshold value of threshold ratings curve have allocated weighted value in advance.
Wherein, scene can be made of following four situation elements namely system, environment, system user distribution and terminal
Type, the embodiment of the present invention are not especially limited this.System refer to verification system type, be specifically as follows vocal print, face,
Fingerprint, iris etc..Environment refer to verification local environment, be specifically as follows hall, square, high speed, road, office, market,
Dining room, classroom etc..System user distribution refers to the statistical information of verification system user, is specifically as follows home and abroad
Statistical information, the statistical information of Regional Distribution, the statistical information of Sex, Age section.Terminal type refers to carrying verification system
The corresponding type of terminal can be mobile phone, computer, tablet etc..
For example, being hall using system as vocal print, environment, identifying system user is distributed as domestic, terminal type as mobile phone
For, which is combined into a kind of scene, and the corresponding sample threshold value appraisal curve of the scene can be made
For first kind sample threshold value appraisal curve.Similarly, system be vocal print, be identified as hall, identifying system user is distributed as
External, terminal type is the sample threshold value appraisal curve corresponding to the scene of mobile phone these four situation elements combination, can
As the second class sample threshold value appraisal curve.It is external, eventually that system is face, environment is that hall, system user are distributed as
The sample threshold value appraisal curve corresponding to the scene that type is mobile phone these four situation elements combination is held, third is can be used as
Class sample threshold value appraisal curve.Above-mentioned three classes sample threshold value appraisal curve is input in disaggregated model and is instructed
Practice, the input of training pattern is set as the curvilinear characteristic of each sample threshold value appraisal curve and each sample thresholding threshold
It is worth the first threshold value and the second threshold value of appraisal curve, the output of training pattern is set as curve category.According to defeated
The curve category gone out can determine the curve category the weighted value of pre-set first threshold value and the second thresholding threshold
The weighted value of value.Specifically, the corresponding classification of sample threshold value appraisal curve and weighted value can refer to such as the following table 1：
Table 1
In table 1, it is enumerated the sample threshold value appraisal curve of five kinds of curve categories.Based on curve discrimination model,
The exportable sample threshold value appraisal curve to match with threshold value appraisal curve.Specifically, curve discrimination model can be defeated
Go out curve category, based on the content in upper table 1, you can determine corresponding to the sample threshold value appraisal curve of arbitrary curve classification
Weighted value.Wherein, sigmoid weights are the corresponding weighted value of the second threshold value, and curve distance weight is first
Limit the corresponding weighted value of threshold value.
By the content of above-described embodiment it is found that the type of curve discrimination model can also be regression model.Correspondingly, right
It, can be by the curvilinear characteristic of each sample threshold value appraisal curve, the first threshold value and second when regression model is trained
Threshold value is as input, using the corresponding weighted value of the first threshold value of each sample threshold value appraisal curve as defeated
Go out.At this point, practical when determining weighted value using curve discrimination model, output is the corresponding weight of the first threshold value
Value.If it is by the curvilinear characteristic of each sample threshold value appraisal curve, the first thresholding threshold when being trained to regression model
Value and the second threshold value are as input, by the corresponding weighted value of the second threshold value of each sample threshold value appraisal curve
As output, then practical when determining weighted value using curve discrimination model, output is the corresponding power of the second threshold value
Weight values.
By the content of above-described embodiment it is found that the embodiment of the present invention, which can be directed to current scene, rationally determines a thresholding threshold
Value, and the threshold value can be applied in many verification systems.Correspondingly, referring to Fig. 2, the embodiment of the present invention additionally provides one kind
The scene verification method of verification system, including：
201, the corresponding score of data to be verified is obtained, score is that data to be verified are input to current scene is corresponding to be tested
It is obtained after card system.
Wherein, data to be verified be under current scene input verification system data, according to verification system Authentication-Type,
The data for inputting verification system can be different.For example, recognition of face verification system input is then human face data, fingerprint recognition is tested
The input of card system is then finger print data, and what iris recognition verified system input is then iris data.Score refers to will be to be verified
After data input carries out verification system, system treats the marking of verify data as a result, can determine whether number to be verified based on score automatically
According to verification can be passed through.Wherein, the code of points of different verification systems may be different, also different to the meaning of score, this hair
Bright embodiment is not especially limited this.
If 202, score is more than threshold value, it is determined that data to be verified pass through verification.
Wherein, threshold value refers to the threshold value under current scene, and the threshold value under current scene can be based on upper
Embodiment is stated to be determined.After data to be verified under current scene are inputted corresponding verification system, verification system can export
One score judges whether the data to be verified can pass through verification according to the comparison of the size of the score and threshold value.
When score is more than threshold value, you can determine that data to be verified pass through verification.Certainly, in actual implementation, it is understood that there may be score
When less than threshold value, it is determined that data to be verified are not especially limited this by the situation of verification, the embodiment of the present invention.
Method provided in an embodiment of the present invention, by the way that the corresponding score of data to be verified is compared with threshold value,
Since threshold value has accuracy and robustness, to which verification accuracy can be improved.
Content based on above-described embodiment, an embodiment of the present invention provides a kind of threshold value determining device, the thresholding thresholds
The threshold value that value determining device is used to execute in above method embodiment determines method.Referring to Fig. 3, which includes：
Acquisition module 301, for obtaining the first threshold value, the first threshold value is to be based on the second threshold value and thresholding
Determined by threshold ratings curve, threshold value appraisal curve is referred to based on the evaluation for influencing scene verification result under current scene
What mark was generated, the second threshold value is based on determined by sigmoid functions and threshold value appraisal curve；
Determining module 302, for being based on the corresponding weighted value of the first threshold value and the corresponding power of the second threshold value
Weight values are weighted averagely, using summed result as final threshold value the first threshold value and the second threshold value.
As a kind of alternative embodiment, acquisition module 301 includes：
Distance determining unit is used for based on threshold value point at a distance from several equalization points are on threshold value appraisal curve
Relationship determines that the first threshold value, threshold value point are that the second threshold value corresponds to the point on threshold value appraisal curve, puts down
Point of contact of the weighing apparatus point between the tangent line and threshold value appraisal curve of default slope.
As a kind of alternative embodiment, distance determining unit, including：
Computation subunit, for calculating threshold value point at a distance from each equalization point is on threshold value appraisal curve；
Determination subelement, for using the corresponding equalization point of the shortest distance as target equilibrium point, target equilibrium point to be first
Threshold value corresponds to the point on threshold value appraisal curve.
As a kind of alternative embodiment, it is 1 to preset slope；Correspondingly, threshold value appraisal curve is ROC curve.
As a kind of alternative embodiment, which further includes：
Output module is used for the curvilinear characteristic of threshold value appraisal curve, the first threshold value and the second threshold value
It is input to curve discrimination model, determines the corresponding weighted value of the first threshold value and the corresponding weighted value of the second threshold value.
As a kind of alternative embodiment, curve discrimination model is by the sample threshold value under different types of historic scenery
Appraisal curve is input to disaggregated model or regression model be trained after obtain.
The device of the embodiment of the present invention, by obtaining the first threshold value and the second threshold value, then by the first thresholding threshold
Value and the second threshold value are weighted averagely, with the threshold value that determination is final.Wherein, the first threshold value is to be based on second
Determined by threshold value and threshold value appraisal curve, and the second threshold value is to be based on sigmoid functions and thresholding threshold
It is worth determined by appraisal curve.Since sigmoid functions can be commented with threshold value during determination final threshold value
Valence curve is combined, to may make output result unique.At the same time, by by the first threshold value and the second thresholding threshold
Value is weighted averagely, and the accuracy and robustness of final threshold value can be improved.
Secondly, by regarding the corresponding equalization point of the shortest distance as target equilibrium point, namely based on apart from nearest balance
Point is modified thresholding threshold point, so that the first threshold value determined is more accurate.
Again, the point of contact of the tangent line and ROC curve that are 1 by using slope ensure that as equalization point at equalization point
The amplitude of variation of misclassification rate and the amplitude of variation of recall rate are consistent, to which misclassification rate and recall rate can reach at equalization point
Active balance.
In addition, by the way that the curvilinear characteristic of threshold value appraisal curve, the first threshold value and the second threshold value are inputted
The sample threshold value appraisal curve to match to curve discrimination model, output with threshold value appraisal curve, then it is based on sample
The corresponding weighted value of threshold value appraisal curve determines that the corresponding weighted value of the first threshold value and the second threshold value correspond to
Weighted value, so as to quickly and accurately determine weighted value.
Finally, by the way that the corresponding score of data to be verified to be compared with threshold value, since threshold value has standard
True property and robustness, to which verification accuracy can be improved.
An embodiment of the present invention provides a kind of threshold values to determine equipment, as shown in figure 4, the equipment includes：Processor
(processor) 401, memory (memory) 402 and bus 403；
Wherein, processor 401 and memory 402 complete mutual communication by bus 403 respectively；Processor 401 is used
In calling the program instruction in memory 402, method is determined to execute the threshold value that above-described embodiment is provided, such as wrap
It includes：Obtain the first threshold value, the first threshold value be based on determined by the second threshold value and threshold value appraisal curve,
Threshold value appraisal curve is generated based on the evaluation index for influencing scene verification result under current scene, the second thresholding threshold
Value is based on determined by sigmoid functions and threshold value appraisal curve；Based on the corresponding weighted value of the first threshold value
And the second corresponding weighted value of threshold value, the first threshold value and the second threshold value are weighted average, will summed
As a result as final threshold value.
The embodiment of the present invention also provides a kind of non-transient computer readable storage medium, the non-transient computer readable storage
Medium storing computer instructs, which makes computer execute the threshold value determination side that corresponding embodiment is provided
Method, such as including：The first threshold value is obtained, the first threshold value is to be based on the second threshold value and threshold value appraisal curve
Identified, threshold value appraisal curve is generated based on the evaluation index for influencing scene verification result under current scene,
Second threshold value is based on determined by sigmoid functions and threshold value appraisal curve；Based on the first threshold value pair
The weighted value and the corresponding weighted value of the second threshold value answered, are weighted the first threshold value and the second threshold value flat
, using summed result as final threshold value.
One of ordinary skill in the art will appreciate that：Realize that all or part of step of above method embodiment can pass through
The relevant hardware of program instruction is completed, and program above-mentioned can be stored in a computer read/write memory medium, the program
When being executed, step including the steps of the foregoing method embodiments is executed；And storage medium above-mentioned includes：ROM, RAM, magnetic disc or light
The various media that can store program code such as disk.
Threshold value described above determines that the embodiments such as equipment are only schematical, wherein being said as separating component
Bright unit may or may not be physically separated, and the component shown as unit can be or can not also
It is physical unit, you can be located at a place, or may be distributed over multiple network units.It can be according to actual need
Some or all of module therein is selected to achieve the purpose of the solution of this embodiment.Those of ordinary skill in the art are not paying
In the case of going out performing creative labour, you can to understand and implement.
Through the above description of the embodiments, those skilled in the art can be understood that each embodiment can
It is realized by the mode of software plus required general hardware platform, naturally it is also possible to pass through hardware.Based on this understanding, on
Stating technical solution, substantially the part that contributes to existing technology can be expressed in the form of software products in other words, should
Computer software product can store in a computer-readable storage medium, such as ROM/RAM, magnetic disc, CD, including several fingers
It enables and using so that a computer equipment (can be personal computer, server or the network equipment etc.) executes each implementation
Certain Part Methods of example or embodiment.
Finally it should be noted that：The above embodiments are merely illustrative of the technical solutions of the present invention, rather than its limitations；Although
Present invention has been described in detail with reference to the aforementioned embodiments, it will be understood by those of ordinary skill in the art that：It still may be used
With technical scheme described in the above embodiments is modified or equivalent replacement of some of the technical features；
And these modifications or replacements, various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution spirit and
Range.
Claims (10)
1. a kind of threshold value determines method, which is characterized in that including：
The first threshold value is obtained, first threshold value is true based on the second threshold value and threshold value appraisal curve institute
Fixed, the threshold value appraisal curve is generated based on the evaluation index for influencing scene verification result under current scene,
Second threshold value is based on determined by sigmoid functions and the threshold value appraisal curve；
Based on the corresponding weighted value of first threshold value and the corresponding weighted value of second threshold value, to described
One threshold value and second threshold value are weighted averagely, using summed result as final threshold value.
2. according to the method described in claim 1, it is characterized in that, the first threshold value of the acquisition, including：
Based on threshold value point and distance relation of several equalization points on the threshold value appraisal curve, described first is determined
Threshold value, the threshold value point are that second threshold value corresponds to the point on the threshold value appraisal curve, institute
State the point of contact between the tangent line and the threshold value appraisal curve that equalization point is default slope.
3. according to the method described in claim 2, it is characterized in that, described be based on threshold value point with several equalization points described
Distance relation on threshold value appraisal curve determines first threshold value, including：
The threshold value point is calculated at a distance from each equalization point is on the threshold value appraisal curve；
Using the corresponding equalization point of the shortest distance as target equilibrium point, the target equilibrium point corresponds to for first threshold value
Point on the threshold value appraisal curve.
4. according to the method in claim 2 or 3, which is characterized in that the default slope is 1；Correspondingly, the thresholding threshold
Value appraisal curve is ROC curve.
5. according to the method described in claim 1, it is characterized in that, described be based on the corresponding weighted value of first threshold value
And the corresponding weighted value of second threshold value, first threshold value and second threshold value are weighted
Averagely, before using summed result as final threshold value, further include：
The curvilinear characteristic of the threshold value appraisal curve, first threshold value and second threshold value are input to
Curve discrimination model determines the corresponding weighted value of first threshold value and the corresponding weight of second threshold value
Value.
6. according to the method described in claim 5, it is characterized in that, the curve discrimination model is by different types of history field
Sample threshold value appraisal curve under scape is input to disaggregated model or regression model be trained after obtain.
7. a kind of scene verification method, which is characterized in that including：
Obtain the corresponding score of data to be verified, the score is that the data to be verified are input to current scene is corresponding to be tested
It is obtained after card system；
If the score is more than threshold value, it is determined that for the data to be verified by verification, the threshold value is by right
It is required that determined by any one of 1 to 6 method.
8. a kind of threshold value determining device, which is characterized in that including：
Acquisition module, for obtaining the first threshold value, first threshold value is to be based on the second threshold value and thresholding threshold
It is worth determined by appraisal curve, the threshold value appraisal curve is based on the evaluation for influencing scene verification result under current scene
What index was generated, second threshold value is determined based on sigmoid functions and the threshold value appraisal curve
's；
Determining module, for being based on the corresponding weighted value of first threshold value and the corresponding power of second threshold value
Weight values are weighted averagely, using summed result as final door first threshold value and second threshold value
Limit threshold value.
9. a kind of threshold value determines equipment, which is characterized in that including：
At least one processor；
And at least one processor being connect with the processor communication, wherein：
The memory is stored with the program instruction that can be executed by the processor, and the processor calls described program to instruct energy
Enough methods executed as described in claim 1 to 6 is any.
10. a kind of non-transient computer readable storage medium, which is characterized in that the non-transient computer readable storage medium is deposited
Computer instruction is stored up, the computer instruction makes the computer execute the method as described in claim 1 to 6 is any.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810421678.5A CN108647874B (en) | 2018-05-04 | 2018-05-04 | Threshold value determining method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810421678.5A CN108647874B (en) | 2018-05-04 | 2018-05-04 | Threshold value determining method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108647874A true CN108647874A (en) | 2018-10-12 |
| CN108647874B CN108647874B (en) | 2020-12-08 |
Family
ID=63749412
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810421678.5A Active CN108647874B (en) | 2018-05-04 | 2018-05-04 | Threshold value determining method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108647874B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070175A (en) * | 2019-04-12 | 2019-07-30 | 北京市商汤科技开发有限公司 | Image processing method, model training method and device, electronic equipment |
| RU2731332C1 (en) * | 2020-01-16 | 2020-09-01 | Акционерное общество "Концерн "Созвездие" | Method for formation of decisionmaking device threshold based on neuroregulator |
| CN111898498A (en) * | 2020-07-16 | 2020-11-06 | 北京市商汤科技开发有限公司 | Matching threshold determination method, identity verification method, device and storage medium |
| CN112597810A (en) * | 2020-06-01 | 2021-04-02 | 支付宝实验室(新加坡)有限公司 | Identity document authentication method and system |
| CN112951247A (en) * | 2021-03-23 | 2021-06-11 | 上海掌数科技有限公司 | Method for quickly verifying voiceprint based on application scene and application thereof |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130204880A1 (en) * | 2012-02-06 | 2013-08-08 | Fis Financial Compliance Solutions, Llc | Methods and systems for list filtering based on known entity matching |
| CN104598795A (en) * | 2015-01-30 | 2015-05-06 | 科大讯飞股份有限公司 | Authentication method and system |
| CN105550677A (en) * | 2016-02-02 | 2016-05-04 | 河北大学 | 3D palm print identification method |
| CN107066983A (en) * | 2017-04-20 | 2017-08-18 | 腾讯科技（上海）有限公司 | A kind of auth method and device |
-
2018
- 2018-05-04 CN CN201810421678.5A patent/CN108647874B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130204880A1 (en) * | 2012-02-06 | 2013-08-08 | Fis Financial Compliance Solutions, Llc | Methods and systems for list filtering based on known entity matching |
| CN104598795A (en) * | 2015-01-30 | 2015-05-06 | 科大讯飞股份有限公司 | Authentication method and system |
| CN105550677A (en) * | 2016-02-02 | 2016-05-04 | 河北大学 | 3D palm print identification method |
| CN107066983A (en) * | 2017-04-20 | 2017-08-18 | 腾讯科技（上海）有限公司 | A kind of auth method and device |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110070175A (en) * | 2019-04-12 | 2019-07-30 | 北京市商汤科技开发有限公司 | Image processing method, model training method and device, electronic equipment |
| CN110070175B (en) * | 2019-04-12 | 2021-07-02 | 北京市商汤科技开发有限公司 | Image processing method, model training method and device and electronic equipment |
| RU2731332C1 (en) * | 2020-01-16 | 2020-09-01 | Акционерное общество "Концерн "Созвездие" | Method for formation of decisionmaking device threshold based on neuroregulator |
| CN112597810A (en) * | 2020-06-01 | 2021-04-02 | 支付宝实验室(新加坡)有限公司 | Identity document authentication method and system |
| CN111898498A (en) * | 2020-07-16 | 2020-11-06 | 北京市商汤科技开发有限公司 | Matching threshold determination method, identity verification method, device and storage medium |
| CN111898498B (en) * | 2020-07-16 | 2024-06-11 | 北京市商汤科技开发有限公司 | Matching threshold determining method, identity verification method, device and storage medium |
| CN112951247A (en) * | 2021-03-23 | 2021-06-11 | 上海掌数科技有限公司 | Method for quickly verifying voiceprint based on application scene and application thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108647874B (en) | 2020-12-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108647874A (en) | Threshold value determines method and device | |
| WO2019205325A1 (en) | Method for determining risk level of user, terminal device, and computer-readable storage medium | |
| CN109241418A (en) | Abnormal user recognition methods and device, equipment, medium based on random forest | |
| CN111080442A (en) | Credit scoring model construction method, device, equipment and storage medium | |
| CN109190351A (en) | On-line signature person identity authorization system based on mobile terminal, device and method | |
| CN111507470A (en) | Abnormal account identification method and device | |
| CN109688275A (en) | Harassing call recognition methods, device and storage medium | |
| CN107995370A (en) | Call control method, device and storage medium and mobile terminal | |
| KR102218506B1 (en) | Account complaint handling method and server | |
| CN107644098A (en) | A kind of fraud recognition methods, device, equipment and storage medium | |
| WO2018006631A1 (en) | User level automatic segmentation method and system | |
| CN110930038A (en) | Loan demand identification method, loan demand identification device, loan demand identification terminal and loan demand identification storage medium | |
| CN110020868A (en) | Anti- fraud module Decision fusion method based on online trading feature | |
| CN109656366A (en) | Emotional state identification method and device, computer equipment and storage medium | |
| CN109242307A (en) | A kind of anti-fraudulent policies analysis method, server, electronic equipment and storage medium | |
| CN111241992A (en) | Face recognition model construction method, recognition method, device, equipment and storage medium | |
| CN104751350B (en) | A kind of method for information display and terminal | |
| CN115034305A (en) | Method, system and storage medium for identifying fraudulent users in a speech network using a human-in-loop neural network | |
| CN109753561B (en) | Automatic reply generation method and device | |
| CN105450412A (en) | Identity authentication method and device | |
| CN114139931A (en) | Enterprise data evaluation method and device, computer equipment and storage medium | |
| CN111404835B (en) | Flow control method, device, equipment and storage medium | |
| CN110210960A (en) | A kind of data adjustment method and relevant device based on data analysis | |
| CN111368131A (en) | User relationship identification method and device, electronic equipment and storage medium | |
| CN116166993A (en) | Power line fault type identification method and device, power system and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |