CN108647874B - Threshold value determining method and device - Google Patents
Threshold value determining method and device Download PDFInfo
- Publication number
- CN108647874B CN108647874B CN201810421678.5A CN201810421678A CN108647874B CN 108647874 B CN108647874 B CN 108647874B CN 201810421678 A CN201810421678 A CN 201810421678A CN 108647874 B CN108647874 B CN 108647874B
- Authority
- CN
- China
- Prior art keywords
- threshold
- threshold value
- curve
- evaluation
- point
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
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
The embodiment of the invention provides a threshold value determining method and device, and belongs to the field of verification and identification. The method comprises the following steps: acquiring a first threshold, wherein the first threshold is determined based on a second threshold and a threshold evaluation curve, the threshold evaluation curve is generated based on evaluation indexes influencing a scene verification result under a current scene, and the second threshold is determined based on a sigmoid function and the threshold evaluation curve; and carrying out weighted average on the first threshold value and the second threshold value, and taking the summation result as a final threshold value. The sigmoid function can be combined with the threshold evaluation curve in the process of determining the final threshold, so that the output result is unique. Meanwhile, the accuracy and robustness of the final threshold value can be improved by carrying out weighted average on the first threshold value and the second threshold value.
Description
Technical Field
The embodiment of the invention relates to the field of verification and identification, in particular to a threshold value determining method and device.
Background
At present, a biometric technology is applied to more and more systems, whether in a card punching system, a shopping payment system or a face-swiping authentication system, a threshold value is usually set, and when the system acquires data to be authenticated, a score corresponding to the data to be authenticated can be calculated. And determining whether the data to be verified can be verified based on the score and a threshold value.
In the related art, a threshold evaluation curve is generally generated according to evaluation indexes affecting the verification result in the system, such as: ROC curve, DET curve, and PR curve. And then, based on a threshold evaluation curve, a unique threshold is directly determined by the sigmoid function, and the accuracy and robustness of the determined threshold are poor.
Disclosure of Invention
To solve the above problems, embodiments of the present invention provide a threshold value determining method and apparatus that overcome the above problems or at least partially solve the above problems.
According to a first aspect of the embodiments of the present invention, there is provided a threshold value determining method, including:
acquiring a first threshold, wherein the first threshold is determined based on a second threshold and a threshold evaluation curve, the threshold evaluation curve is generated based on evaluation indexes influencing a scene verification result under a current scene, and the second threshold is determined based on a sigmoid function and the threshold evaluation curve;
and carrying out weighted average on the first threshold and the second threshold based on the weight value corresponding to the first threshold and the weight value corresponding to the second threshold, and taking the summation result as the final threshold.
According to the method provided by the embodiment of the invention, the final threshold value is determined by obtaining the first threshold value and the second threshold value and then carrying out weighted average on the first threshold value and the second threshold value. The first threshold value is determined based on a second threshold value and a threshold value evaluation curve, and the second threshold value is determined based on a sigmoid function and the threshold value evaluation curve. The sigmoid function can be combined with the threshold evaluation curve in the process of determining the final threshold, so that the output result is unique. Meanwhile, the accuracy and robustness of the final threshold value can be improved by carrying out weighted average on the first threshold value and the second threshold value.
According to a second aspect of the embodiments of the present invention, there is provided a threshold value determining apparatus, including:
the device comprises an acquisition module, a verification module and a comparison module, wherein the acquisition module is used for acquiring a first threshold, the first threshold is determined based on a second threshold and a threshold evaluation curve, the threshold evaluation curve is generated based on evaluation indexes influencing a scene verification result under a current scene, and the second threshold is determined based on a sigmoid function and the threshold evaluation curve;
and the determining module is used for carrying out weighted average on the first threshold value and the second threshold value based on the weight value corresponding to the first threshold value and the weight value corresponding to the second threshold value, and taking the summation result as the final threshold value.
According to a third aspect of the embodiments of the present invention, there is provided a scene verification method, including:
obtaining a score corresponding to the data to be verified, wherein the score is obtained after the data to be verified is input to a verification system corresponding to a current scene;
and if the score is larger than the threshold value, determining that the data to be verified passes the verification, wherein the threshold value is determined by the threshold value determining method.
According to a fourth aspect of the embodiments of the present invention, there is provided a threshold value determining device including:
at least one processor; and
at least one memory communicatively coupled to the processor, wherein:
the memory stores program instructions executable by the processor, the processor invoking the program instructions to perform the threshold determination method provided by any of the various possible implementations of the first aspect.
According to a fifth aspect of embodiments of the present invention, there is provided a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the method for determining a threshold value provided in any one of the various possible implementations of the first aspect.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory and are not restrictive of embodiments of the invention.
Drawings
Fig. 1 is a schematic flow chart of a threshold value determining method according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of a method for scene verification according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a threshold value determining apparatus according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a threshold value determining apparatus according to an embodiment of the present invention.
Detailed Description
The following describes embodiments of the present invention in further detail with reference to the drawings and examples. The following examples are intended to illustrate the examples of the present invention, but are not intended to limit the scope of the examples of the present invention.
In the related art, the threshold value is generally determined in two ways. One way is to obtain positive and negative example scene data with the appropriate amount of accuracy requirement according to the scene used by the system and the accuracy requirement of the data to be verified. Inputting the positive and negative scene data into a verification system, outputting corresponding evaluation indexes under each score threshold, drawing a threshold evaluation curve according to the evaluation indexes, and taking the position of a balance point in the threshold evaluation curve as a final threshold. And the other mode is that a sigmoid function is used for regulating a threshold evaluation curve obtained by drawing the evaluation index to a smooth threshold curve, and then a parameter value is configured in advance by the sigmoid function to output a final threshold.
However, in the two existing schemes, the first scheme directly adopts the position of the balance point in the threshold evaluation curve as the final threshold, and when there are multiple positions of the balance point in the threshold evaluation curve, the system cannot judge which balance point can be used as the final threshold, or the accuracy of the judged final threshold is low, which is not in line with the actual requirement. In the second scheme, a sigmoid function is adopted to directly output a final threshold value, the robustness is poor in the using process of the system, manual testing, calculation and adjustment are needed again aiming at each different scene, and the processing process of the system is complicated.
In view of the above situation, an embodiment of the present invention provides a threshold value determining method. Referring to fig. 1, the method includes:
101. the method comprises the steps of obtaining a first threshold value, wherein the first threshold value is determined based on a second threshold value and a threshold value evaluation curve, the threshold value evaluation curve is generated based on evaluation indexes influencing a scene verification result under a current scene, and the second threshold value is determined based on a sigmoid function and the threshold value evaluation curve.
In step 101, a threshold evaluation curve is generated based on evaluation indexes affecting a scene verification result in a current scene, where the evaluation indexes affecting the scene verification result may be a false recognition rate and a recall rate. Each threshold value corresponds to a group of false recognition rate and recall rate and can be respectively used as an abscissa and an ordinate. And (4) drawing a threshold evaluation curve according to the false recognition rate and the recall rate corresponding to the multiple threshold values. The false recognition rate and the recall rate corresponding to each threshold value can be obtained through a large amount of data statistics. For any threshold, the false recognition rate represents the statistical frequency that the verification data which should not pass under the condition of the threshold passes, and can be represented by the following formula:
wherein, FAR is the false recognition rate, NFA is the number of times that the NFA should not pass the verification but pass the verification during the verification, and NIRA is the total number of times of the verification. The recall rate represents the ratio of the number of passed verifications to the total number of verifications under the threshold condition.
It will be appreciated that the higher the false positive rate requirement of the authentication system, the higher the security requirement of the authentication system is justified. For example, a less secure authentication system for entertainment may require only a percent of the false positive, while an authentication system for security, financial, etc. applications may require one thousandth or one ten thousandth of the false positive. Of course, besides the error recognition rate and the recall rate as the evaluation indexes, other evaluation indexes such as the accuracy and the error rate may also be adopted, which is not specifically limited in the embodiment of the present invention. For convenience of description, the present invention is described by taking the error recognition rate and the recall rate as evaluation indexes.
The second threshold value is determined based on the sigmoid function and a threshold value evaluation curve. Specifically, a sigmoid function is used for regulating a threshold evaluation curve into a smooth interval, and then a unique threshold is output and used as a second threshold according to a parameter value preset by the sigmoid. Wherein the smooth interval can take the value [0, 1%]The standard function of the sigmoid function is:
considering that the whole threshold evaluation curve can not be regulated to a smooth interval, the embodiment of the invention adds two parameters a and b for translation and stretching on the basis of the standard function to obtain the transformed sigmoid function:
two points are selected according to requirements in the curve: (y)1，y1′)、(y2，y2') and substitute these two points into
And calculating to obtain a value and b value. And finally, adjusting the value of y' according to the actual requirement, thereby calculating the value of y and taking the value of y as a second threshold value. Because the regular smooth interval is [0,1 ]]Therefore, the value of y' is preferably set to 0.6, which means that 60 points are reached, and the verification data can be considered to pass. Of course, the specific value of y' may be adjusted according to actual conditions, and this is not specifically limited in the embodiment of the present invention.
102. And carrying out weighted average on the first threshold and the second threshold based on the weight value corresponding to the first threshold and the weight value corresponding to the second threshold, and taking the summation result as the final threshold.
In step 102, the weight value corresponding to the first threshold and the weight value corresponding to the second threshold may be set according to requirements, which is not specifically limited in the embodiment of the present invention. When the first threshold is more important, the weight ratio allocated to the first threshold may be increased, and when the evaluation of the second threshold is more important, the weight ratio allocated to the second threshold may be increased, and the sum of the weight value corresponding to the first threshold and the weight value corresponding to the second threshold is 1. Specifically, in some conventional scenarios with less change, such as scenes of fingerprint recognition and face recognition for moving up and down every day, since the scenes are relatively fixed and have less change, the result of the second threshold calculation can accurately reflect the verification result, so that the weight ratio of the second threshold can be set to 0.8 or 0.9, or even can be directly set to 1. In some scenes with complex changes, such as iris recognition scenes of bank safes, since the scenes are used less and the safety performance requirement is higher, the result of calculation by using the second threshold value alone cannot be used as the final threshold value, so that the weight ratio of the first threshold value can be increased according to the actual situation, for example, the weight ratio of the first threshold value is set to 0.6 or 0.7. In addition, when the weight ratio cannot be reasonably distributed according to the actual situation in some scenarios, both the weight ratio of the first threshold and the weight ratio of the second threshold may be set to 0.5, which is not specifically limited in the embodiment of the present invention.
According to the method provided by the embodiment of the invention, the final threshold value is determined by obtaining the first threshold value and the second threshold value and then carrying out weighted average on the first threshold value and the second threshold value. The first threshold value is determined based on a second threshold value and a threshold value evaluation curve, and the second threshold value is determined based on a sigmoid function and the threshold value evaluation curve. The sigmoid function can be combined with the threshold evaluation curve in the process of determining the final threshold, so that the output result is unique. Meanwhile, the accuracy and robustness of the final threshold value can be improved by carrying out weighted average on the first threshold value and the second threshold value.
As can be seen from the content of the above embodiment, the embodiment of the present invention can determine the second threshold based on the sigmoid function and the threshold evaluation curve, and can determine the first threshold based on the second threshold and the threshold evaluation curve. Based on the content of the foregoing embodiment, as an optional embodiment, the embodiment of the present invention does not specifically limit the manner of obtaining the first threshold, which includes but is not limited to: and determining a first threshold value based on the distance relationship between the threshold value point and a plurality of balance points on the threshold value evaluation curve, wherein the threshold value point is a point of a second threshold value corresponding to the threshold value evaluation curve, and the balance point is a tangent point between a tangent line of a preset slope and the threshold value evaluation curve.
And the threshold value point is a point of the second threshold value corresponding to the threshold value evaluation curve. As can be seen from the content of the above embodiment, in the embodiment of the present invention, the second threshold value may be output on the threshold evaluation curve through the sigmoid function, and the coordinate of the second threshold value on the threshold evaluation curve corresponding to the threshold is the threshold point. The balance point is a tangent point between a tangent line of the preset slope and the threshold evaluation curve, and the tangent line of the preset slope can be a plurality of parallel straight lines drawn under the same coordinate system corresponding to the threshold evaluation curve. If a tangent point exists between any tangent line with a preset slope and the threshold evaluation curve, the tangent point can be used as an equilibrium point.
According to the method provided by the embodiment of the invention, the first threshold value is determined based on the distance relationship between the threshold value point and a plurality of balance points on the threshold value evaluation curve. Because there may be a plurality of balance points, based on the distance relationship between the threshold point and the balance point on the threshold evaluation curve, the point closest to the threshold point in the threshold value can be determined, so that the determined first threshold value is more accurate.
Based on the content of the foregoing embodiment, as an optional embodiment, the embodiment of the present invention does not specifically limit the manner of determining the first threshold based on the distance relationship between the threshold point and the plurality of balance points on the threshold evaluation curve, including but not limited to: calculating the distance between a threshold point and each balance point on a threshold evaluation curve; and taking the balance point corresponding to the shortest distance as a target balance point, wherein the target balance point is a point on the threshold evaluation curve corresponding to the first threshold.
The target balance point represents a point on a threshold evaluation curve corresponding to the first threshold, and the point on the threshold evaluation curve and the threshold have a corresponding relation, so that the first threshold can be obtained based on the target balance point. It will be appreciated that the distance between the equilibrium point and the threshold point can reflect the proximity of each equilibrium point to the threshold point, with closer equilibrium points demonstrating less variance between the equilibrium point and the threshold point. In the embodiment of the invention, the closer balance point is selected, and the threshold point can be corrected based on the closest balance point on the premise of ensuring the approaching of the real verification result. Thereby making the determined first threshold value more accurate.
According to the method provided by the embodiment of the invention, the balance point corresponding to the shortest distance is used as the target balance point, namely, the threshold value point is corrected based on the balance point with the closest distance, so that the determined first threshold value is more accurate.
Considering that the threshold evaluation curves are of various types, the threshold evaluation curves drawn by each threshold evaluation curve are different according to different evaluation indexes. In addition, the tangent slopes used for determining the balance point for each threshold evaluation curve are different. For example, the slope of the tangent line used when determining the equilibrium point using the PR curve is-1. Based on the content of the above embodiment, as an alternative embodiment, the preset slope is 1; accordingly, the threshold evaluation curve is an ROC curve.
The embodiment of the invention provides a mode for determining a balance point based on a tangent line with a preset slope of 1 and a threshold evaluation curve. In the ROC curve provided by the embodiment of the present invention, the physical meaning represented by the abscissa is the false recognition rate, the physical meaning represented by the ordinate is the recall rate, the physical meaning of the balance point on the ROC curve is that the false recognition rate and the recall rate at the balance point can reach effective balance, and the variation range of the false recognition rate and the variation range of the recall rate are substantially consistent. Accordingly, to keep the variation range of the misrecognition rate and the variation range of the recall rate consistent, the slope of the tangent line representing the point on the ROC curve is equal to one.
According to the method provided by the embodiment of the invention, the tangent line with the slope of 1 and the tangent point of the ROC curve are used as the balance point, so that the change amplitude of the false recognition rate and the change amplitude of the recall rate at the balance point are kept consistent, and the false recognition rate and the recall rate at the balance point can be effectively balanced.
Considering that the first threshold and the second threshold are related to different scenarios when assigning the weight values, and the scenario features have repeatability, if the weight value corresponding to the first threshold and the weight value corresponding to the second threshold need to be determined according to a specific scenario during each verification, a large amount of processing time will be wasted, and the optimal weight value may not necessarily be obtained. Based on the content of the foregoing embodiment, as an optional embodiment, the performing weighted average on the first threshold and the second threshold based on the weight value corresponding to the first threshold and the weight value corresponding to the second threshold, and before taking the summation result as the final threshold, further includes: and inputting the curve characteristics of the threshold evaluation curve, the first threshold and the second threshold into a curve discrimination model, and determining the weight value corresponding to the first threshold and the weight value corresponding to the second threshold.
The curve characteristics of the threshold evaluation curve are characteristics reflecting the variation trend and the variation amplitude of the curve, and preferably, the curve characteristics of each threshold evaluation curve can be represented in the form of a mathematical expression containing a plurality of parameters. The curve discrimination model is a model trained in advance and used for determining the weight value, and specifically may adopt a CNN neural network, an RNN neural network or an SVM classifier, which is not specifically limited in this embodiment of the present invention. It should be noted that the type of the curve discrimination model may be a classification model or a regression model, which is not specifically limited in this embodiment of the present invention. When the curve discrimination model is a classification model, the curve characteristics of the threshold evaluation curve corresponding to the current scene, the first threshold and the second threshold are input into the curve discrimination model, and the category corresponding to the sample threshold evaluation curve matched with the threshold evaluation curve can be output. Before the above process is executed, sample threshold evaluation curves of multiple categories may be preset, and the sample threshold evaluation curve corresponding to the category output by the curve discrimination model is the closest curve to the threshold evaluation curve corresponding to the current scene in the curve discrimination model.
After the class corresponding to the sample threshold evaluation curve is obtained, a weight value corresponding to the first threshold and a weight value corresponding to the second threshold may be determined based on the class corresponding to the sample threshold evaluation curve. Specifically, a set of weight values may be set in advance for the sample threshold evaluation curve for each category. After the curve discrimination model outputs the category corresponding to the sample threshold evaluation curve, a group of weight values corresponding to the category can be respectively used as the weight value corresponding to the first threshold and the weight value corresponding to the second threshold. The group of weight values corresponding to the sample threshold evaluation curve of each category comprises two weight values, the corresponding relation between the two weight values and the first threshold and the second threshold is predetermined, and the sum of the two weight values is 1.
When the curve discrimination model is a regression model, the curve characteristics of the threshold evaluation curve corresponding to the current scene, the first threshold and the second threshold are input to the curve discrimination model, and the weight value corresponding to the first threshold or the weight value corresponding to the second threshold can be output. Since the sum of the weight value corresponding to the first threshold and the weight value corresponding to the second threshold is 1, the weight value corresponding to the first threshold or the weight value corresponding to the second threshold can be obtained at last no matter the weight value corresponding to the first threshold or the weight value corresponding to the second threshold is output.
According to the method provided by the embodiment of the invention, the curve characteristics of the threshold evaluation curve, the first threshold and the second threshold are input into the curve discrimination model, and the weight value corresponding to the first threshold and the weight value corresponding to the second threshold are determined, so that the weight values can be determined quickly and accurately.
As can be seen from the content of the foregoing embodiments, the embodiment of the present invention provides a curve discrimination model, which can directly output the curve characteristics of the input threshold evaluation curve, the first threshold and the second threshold as the sample threshold evaluation curve matched with the threshold evaluation curve, so that the curve discrimination model can be trained in advance. Based on the content of the above embodiment, as an optional embodiment, the curve discrimination model is obtained by inputting the sample threshold evaluation curves in different types of historical scenes into a classification model or a regression model for training.
Specifically, a large amount of historical data of different types of scenes can be obtained in advance, the historical data are stored in the form of threshold evaluation curves, and the stored historical data are used as sample threshold evaluation curves. Further, the sample threshold evaluation curves are classified according to different scene characteristics corresponding to the sample threshold evaluation curves. Each type of sample threshold evaluation curve corresponds to a possible scene, and a first threshold and a second threshold of each type of sample threshold evaluation curve are pre-assigned with weight values.
The scene may be composed of the following four scene elements, that is, the system, the environment, the distribution of system users, and the terminal type, which is not specifically limited in the embodiment of the present invention. The system refers to a verification system type, and specifically can be voiceprint, human face, fingerprint, iris and the like. The environment refers to the environment where the verification is located, and specifically may be a hall, a square, a high speed, a road, an office, a mall, a restaurant, a classroom, and the like. The system user distribution refers to verifying statistical information of system users, and specifically can be statistical information of domestic and foreign countries, statistical information of regional distribution and statistical information of gender and age groups. The terminal type refers to a type corresponding to a terminal carrying the verification system, and can be a mobile phone, a computer, a tablet and the like.
For example, taking the system as a voiceprint, the environment as a hall, the identification system users distributed as domestic, and the terminal type as a mobile phone as an example, the four scene elements can be combined into one scene, and the sample threshold evaluation curve corresponding to the scene can be used as the first type of sample threshold evaluation curve. Similarly, the sample threshold evaluation curve corresponding to the scene in which the system is voiceprint, the system is identified as a hall, the system users are identified as foreign, and the terminal type is the combination of the four scene elements of the mobile phone can be used as a second type of sample threshold evaluation curve. The system is a sample threshold evaluation curve corresponding to a scene of four scene element combinations, namely a human face, a hall environment, a foreign user distribution and a mobile phone terminal type, and can be used as a third type sample threshold evaluation curve. And inputting the three types of sample threshold evaluation curves into a classification model for training, wherein the input of the training model is set as the curve characteristic of each sample threshold evaluation curve and the first threshold and the second threshold of each sample threshold evaluation curve, and the output of the training model is set as the curve type. According to the output curve type, the weight value of a first threshold value and the weight value of a second threshold value preset by the curve type can be determined. Specifically, the category and the weight value corresponding to the sample threshold evaluation curve may refer to the following table 1:
TABLE 1
In table 1, sample threshold evaluation curves for the five curve classes are listed. Based on the curve discrimination model, a sample threshold evaluation curve matched with the threshold evaluation curve can be output. Specifically, the curve discrimination model may output a curve type, and based on the content in table 1 above, a weight value corresponding to a sample threshold evaluation curve of any curve type may be determined. The sigmoid weight is a weight value corresponding to the second threshold, and the curve distance weight is a weight value corresponding to the first threshold.
As can be seen from the above embodiments, the type of the curve discrimination model may also be a regression model. Correspondingly, when the regression model is trained, the curve feature, the first threshold value and the second threshold value of each sample threshold value evaluation curve can be used as input, and the weight value corresponding to the first threshold value of each sample threshold value evaluation curve can be used as output. At this time, when the curve discrimination model is actually used to determine the weight value, the output value is the weight value corresponding to the first threshold value. If the curve feature, the first threshold value and the second threshold value of each sample threshold value evaluation curve are used as input and the weight value corresponding to the second threshold value of each sample threshold value evaluation curve is used as output when the regression model is trained, the output is the weight value corresponding to the second threshold value when the curve discrimination model is used to determine the weight value actually.
From the content of the above embodiments, the embodiments of the present invention can reasonably determine a threshold value for a current scenario, and the threshold value can be applied to many verification systems. Correspondingly, referring to fig. 2, an embodiment of the present invention further provides a scene verification method for a verification system, including:
201. and obtaining a score corresponding to the data to be verified, wherein the score is obtained after the data to be verified is input into a verification system corresponding to the current scene.
The data to be verified is data input into the verification system in the current scene, and the data input into the verification system can be different according to the verification type of the verification system. For example, the face recognition verification system inputs face data, the fingerprint recognition verification system inputs fingerprint data, and the iris recognition verification system inputs iris data. The scoring refers to that after the data to be verified is input into the verification system, the system automatically scores the result of the data to be verified, and the data to be verified can be judged to pass the verification based on the scoring. The scoring rules of different verification systems may be different, so that the scores have different meanings, and this is not particularly limited in the embodiment of the present invention.
202. And if the score is larger than the threshold value, determining that the data to be verified passes the verification.
The threshold value refers to a threshold value in the current scene, and the threshold value in the current scene may be determined based on the above embodiments. After the data to be verified under the current scene is input into the corresponding verification system, the verification system outputs a score, and whether the data to be verified can pass the verification is judged according to the comparison between the score and the threshold value. And when the score is larger than the threshold value, determining that the data to be verified passes the verification. Of course, in actual implementation, when the score may be smaller than the threshold, it is determined that the data to be verified passes the verification, and this is not specifically limited in the embodiment of the present invention.
According to the method provided by the embodiment of the invention, the score corresponding to the data to be verified is compared with the threshold value, and the threshold value has accuracy and robustness, so that the verification accuracy can be improved.
Based on the content of the foregoing embodiments, an embodiment of the present invention provides a threshold value determining apparatus, where the threshold value determining apparatus is configured to execute the threshold value determining method in the foregoing method embodiments. Referring to fig. 3, the apparatus includes:
an obtaining module 301, configured to obtain a first threshold, where the first threshold is determined based on a second threshold and a threshold evaluation curve, the threshold evaluation curve is generated based on an evaluation index affecting a scene verification result in a current scene, and the second threshold is determined based on a sigmoid function and the threshold evaluation curve;
the determining module 302 is configured to perform weighted average on the first threshold and the second threshold based on the weight value corresponding to the first threshold and the weight value corresponding to the second threshold, and use the summation result as a final threshold.
As an alternative embodiment, the obtaining module 301 includes:
and the distance determining unit is used for determining a first threshold value based on the distance relationship between the threshold value point and a plurality of balance points on the threshold value evaluation curve, wherein the threshold value point is a point of a second threshold value corresponding to the threshold value evaluation curve, and the balance points are tangent points between a tangent line of a preset slope and the threshold value evaluation curve.
As an alternative embodiment, the distance determination unit comprises:
the calculating subunit is used for calculating the distance between the threshold point and each balance point on the threshold evaluation curve;
and the determining subunit is used for taking the balance point corresponding to the shortest distance as a target balance point, wherein the target balance point is a point on the threshold evaluation curve corresponding to the first threshold.
As an alternative embodiment, the preset slope is 1; accordingly, the threshold evaluation curve is an ROC curve.
As an alternative embodiment, the apparatus further comprises:
and the output module is used for inputting the curve characteristics of the threshold evaluation curve, the first threshold and the second threshold into the curve discrimination model and determining the weight value corresponding to the first threshold and the weight value corresponding to the second threshold.
As an optional embodiment, the curve discrimination model is obtained by inputting the sample threshold evaluation curves in different types of historical scenes into a classification model or a regression model for training.
The device of the embodiment of the invention determines the final threshold value by obtaining the first threshold value and the second threshold value and then carrying out weighted average on the first threshold value and the second threshold value. The first threshold value is determined based on a second threshold value and a threshold value evaluation curve, and the second threshold value is determined based on a sigmoid function and the threshold value evaluation curve. The sigmoid function can be combined with the threshold evaluation curve in the process of determining the final threshold, so that the output result is unique. Meanwhile, the accuracy and robustness of the final threshold value can be improved by carrying out weighted average on the first threshold value and the second threshold value.
Secondly, the balance point corresponding to the shortest distance is used as a target balance point, namely, the threshold point is corrected based on the balance point with the shortest distance, so that the determined first threshold is more accurate.
And thirdly, by using the tangent line with the slope of 1 and the tangent point of the ROC curve as a balance point, the change range of the misrecognition rate and the change range of the recall rate at the balance point are ensured to be consistent, so that the misrecognition rate and the recall rate at the balance point can be effectively balanced.
In addition, the curve characteristics of the threshold evaluation curve, the first threshold and the second threshold are input into the curve discrimination model, the sample threshold evaluation curve matched with the threshold evaluation curve is output, and then the weight value corresponding to the first threshold and the weight value corresponding to the second threshold are determined based on the weight value corresponding to the sample threshold evaluation curve, so that the weight values can be determined quickly and accurately.
And finally, comparing the score corresponding to the data to be verified with a threshold, wherein the threshold has accuracy and robustness, so that the verification accuracy can be improved.
An embodiment of the present invention provides a threshold value determining device, as shown in fig. 4, where the device includes: a processor (processor)401, a memory (memory)402, and a bus 403;
the processor 401 and the memory 402 respectively complete communication with each other through the bus 403; processor 401 is configured to call program instructions in memory 402 to perform the threshold determination method provided by the above embodiments, including: acquiring a first threshold, wherein the first threshold is determined based on a second threshold and a threshold evaluation curve, the threshold evaluation curve is generated based on evaluation indexes influencing a scene verification result under a current scene, and the second threshold is determined based on a sigmoid function and the threshold evaluation curve; and carrying out weighted average on the first threshold and the second threshold based on the weight value corresponding to the first threshold and the weight value corresponding to the second threshold, and taking the summation result as the final threshold.
Embodiments of the present invention further provide a non-transitory computer-readable storage medium storing computer instructions, where the computer instructions cause a computer to execute the threshold determining method provided in the corresponding embodiments, for example, the method includes: acquiring a first threshold, wherein the first threshold is determined based on a second threshold and a threshold evaluation curve, the threshold evaluation curve is generated based on evaluation indexes influencing a scene verification result under a current scene, and the second threshold is determined based on a sigmoid function and the threshold evaluation curve; and carrying out weighted average on the first threshold and the second threshold based on the weight value corresponding to the first threshold and the weight value corresponding to the second threshold, and taking the summation result as the final threshold.
Those of ordinary skill in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as ROM, RAM, magnetic or optical disks.
The embodiments of the threshold determining device and the like described above are merely illustrative, and units illustrated as separate components may or may not be physically separate, and components displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute the various embodiments or some parts of the methods of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (6)
1. A method for scene verification, comprising:
obtaining a score corresponding to data to be verified, wherein the score is obtained after the data to be verified is input to a verification system corresponding to a current scene;
if the score is larger than a threshold value, determining that the data to be verified passes verification;
the threshold is determined by:
acquiring a first threshold, wherein the first threshold is determined based on a second threshold and a threshold evaluation curve, the threshold evaluation curve is generated based on evaluation indexes influencing a scene verification result under a current scene, and the second threshold is determined based on a sigmoid function and the threshold evaluation curve;
and carrying out weighted average on the first threshold value and the second threshold value based on the weight value corresponding to the first threshold value and the weight value corresponding to the second threshold value, and taking a summation result as a final threshold value.
2. The method of claim 1, wherein obtaining the first threshold comprises:
and determining the first threshold value based on the distance relationship between a threshold value point and a plurality of balance points on the threshold value evaluation curve, wherein the threshold value point is a point of the second threshold value on the threshold value evaluation curve corresponding to the threshold value, and the balance point is a tangent point between a tangent line of a preset slope and the threshold value evaluation curve.
3. The method of claim 2, wherein determining the first threshold based on a distance relationship between a threshold point and a number of equilibrium points on the threshold evaluation curve comprises:
calculating the distance between the threshold point and each balance point on the threshold evaluation curve;
and taking the balance point corresponding to the shortest distance as a target balance point, wherein the target balance point is a point on the threshold evaluation curve corresponding to the first threshold.
4. A method according to claim 2 or 3, wherein the preset slope is 1; accordingly, the threshold evaluation curve is an ROC curve.
5. The method according to claim 1, wherein before taking the sum as the final threshold value, the weighted average of the first threshold value and the second threshold value based on the weight value corresponding to the first threshold value and the weight value corresponding to the second threshold value further comprises:
and inputting the curve characteristics of the threshold evaluation curve, the first threshold and the second threshold into a curve discrimination model, and determining a weight value corresponding to the first threshold and a weight value corresponding to the second threshold.
6. The method according to claim 5, wherein the curve discrimination model is obtained by inputting sample threshold evaluation curves in different types of historical scenes into a classification model or a regression model for training.
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 CN108647874A (en) | 2018-10-12 |
| CN108647874B true 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) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| 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 |
| 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 (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9058380B2 (en) * | 2012-02-06 | 2015-06-16 | Fis Financial Compliance Solutions, Llc | Methods and systems for list filtering based on known entity matching |
-
2018
- 2018-05-04 CN CN201810421678.5A patent/CN108647874B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108647874A (en) | 2018-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108647874B (en) | Threshold value determining method and device | |
| CN108765131B (en) | Micro-expression-based credit auditing method, device, terminal and readable storage medium | |
| US20210050020A1 (en) | Voiceprint recognition method, model training method, and server | |
| CN109472206B (en) | Risk assessment method, device, equipment and medium based on micro-expressions | |
| CN112329619B (en) | Face recognition method and device, electronic equipment and readable storage medium | |
| CN109598281B (en) | Service risk prevention and control method, device and equipment | |
| US20170270526A1 (en) | Machine learning for fraud detection | |
| TW202026984A (en) | User identity verification method, device and system | |
| US20150286857A1 (en) | Apparatus and method for recognizing image, and method for generating morphable face images from original image | |
| US20210256208A1 (en) | Method and system for processing bullet screen messages | |
| WO2020211387A1 (en) | Electronic contract displaying method and apparatus, electronic device, and computer readable storage medium | |
| WO2014082496A1 (en) | Method and apparatus for identifying client characteristic and storage medium | |
| CN104427109A (en) | Method for establishing contact item by voices and electronic equipment | |
| US20180233151A1 (en) | Identity vector processing method and computer device | |
| KR20180041727A (en) | Credit Score Model Training Method, Credit Score Calculation Method, Device and Server | |
| WO2019242627A1 (en) | Data processing method and apparatus | |
| WO2023050652A1 (en) | Text recognition-based method for determining esg index in region, and related product | |
| CN111798047A (en) | Wind control prediction method and device, electronic equipment and storage medium | |
| US11899765B2 (en) | Dual-factor identification system and method with adaptive enrollment | |
| CN108831487B (en) | Voiceprint recognition method, electronic device and computer-readable storage medium | |
| CN111552865A (en) | User interest portrait method and related equipment | |
| CN113593579A (en) | Voiceprint recognition method and device and electronic equipment | |
| CN110874567B (en) | Color value judging method and device, electronic equipment and storage medium | |
| WO2021257000A1 (en) | Cross-modal speaker verification | |
| CN110213210B (en) | Network operation request method, server 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 |