U.S. patent application number 17/104578 was filed with the patent office on 2021-05-20 for machine learning for document authentication.
The applicant listed for this patent is Jumio Corporation. Invention is credited to Attila BALOGH, Reinhard HOCHRIESER, Radu ROGOJANU.
Application Number | 20210152550 17/104578 |
Document ID | / |
Family ID | 1000005370578 |
Filed Date | 2021-05-20 |
![](/patent/app/20210152550/US20210152550A1-20210520-D00000.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00001.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00002.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00003.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00004.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00005.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00006.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00007.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00008.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00009.png)
![](/patent/app/20210152550/US20210152550A1-20210520-D00010.png)
View All Diagrams
United States Patent
Application |
20210152550 |
Kind Code |
A1 |
BALOGH; Attila ; et
al. |
May 20, 2021 |
MACHINE LEARNING FOR DOCUMENT AUTHENTICATION
Abstract
Computer systems and methods are provided for using a machine
learning system to analyze authentication information. First
authentication information for a first transaction includes at
least a first image that corresponds to a first identification
document is received. First validation information that corresponds
to a first validation fault is received from a validation system.
Data storage of a machine learning system stores the first
validation information. Second authentication information for a
second transaction includes a second image that corresponds to a
second image is received. The machine learning system determines a
first validation value that corresponds to a probability that the
second image includes the first validation fault. The first
validation value is used to determine whether fault review criteria
are met. In accordance with a determination that the fault review
criteria are met, the second image is transmitted to the validation
system.
Inventors: |
BALOGH; Attila; (Vienna,
AT) ; HOCHRIESER; Reinhard; (Linz, AT) ;
ROGOJANU; Radu; (Vienna, AT) |
|
Applicant: |
Name |
City |
State |
Country |
Type |
Jumio Corporation |
Palo Alto |
CA |
US |
|
|
Family ID: |
1000005370578 |
Appl. No.: |
17/104578 |
Filed: |
November 25, 2020 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
15993366 |
May 30, 2018 |
10880299 |
|
|
17104578 |
|
|
|
|
Current U.S.
Class: |
1/1 |
Current CPC
Class: |
G06N 20/00 20190101;
H04L 63/0861 20130101; G06K 9/00483 20130101 |
International
Class: |
H04L 29/06 20060101
H04L029/06; G06N 20/00 20060101 G06N020/00; G06K 9/00 20060101
G06K009/00 |
Claims
1. (canceled)
2. A computer-implemented method, comprising: at a server system
including one or more processors and memory storing one or more
programs for execution by the one or more processors: receiving,
over a network from a user device remote from the server system, a
first authentication request to authenticate an identity of a user,
wherein the first authentication request includes at least a first
image that corresponds to a first identification document;
determining, using a machine learning system configured to detect
one or more validation faults of a plurality of fault types, a
confidence value that indicates a probability that the first image
includes a first validation fault of the plurality of fault types;
determining whether the confidence value exceeds a threshold
uncertainty value; and in accordance with a determination that the
confidence value exceeds the threshold uncertainty value,
transmitting, to a validation system, the first image.
3. The computer-implemented method of claim 2, wherein the first
validation fault of the plurality of fault types identifies a
corresponding fault location.
4. The computer-implemented method of claim 2, wherein: in
accordance with a determination that the confidence value does not
exceed the threshold uncertainty value, transmitting the first
authentication request to a remote device.
5. The computer-implemented method of claim 2, wherein: in
accordance with a determination that the confidence value exceeds
the threshold uncertainty value: determining whether prompting
criteria are met; and, in accordance with a determination that the
prompting criteria are met, transmitting information for displaying
a prompt for information that corresponds to the first validation
fault of the plurality of fault types in the first identification
document.
6. The computer-implemented method of claim 2, including:
determining, using the machine learning system, a second validation
value that corresponds to a probability that the first image of the
first identification document includes a second validation fault of
the plurality of fault types, distinct from the first validation
fault of the plurality of fault types.
7. The computer-implemented method of claim 6, including:
analyzing, using the machine learning system, a set of incoming
authentication requests; and determining, by the machine learning
system, a subset of the set of incoming authentication requests
that correspond to invalid authentication request activity, wherein
the second validation fault of the plurality of fault types is an
invalidity characteristic of the subset of the set of incoming
authentication requests.
8. The computer-implemented method of claim 2, including:
generating, using the machine learning system, a set of test
authentication requests, wherein each test authentication request
of the set of test authentication requests includes a respective
test validation fault of the plurality of fault types; analyzing,
by the machine learning system, the set of test authentication
requests to generate a set of validation values that corresponds to
the set of test authentication requests; determining that a
respective validation value of the set of validation values does
not meet fault review criteria; and in accordance with the
determination that the respective validation value of the set of
validation values does not meet the fault review criteria,
adjusting the machine learning system using the respective test
validation fault of the plurality of fault types that corresponds
to the respective validation value.
9. A non-transitory computer readable storage medium storing one or
more programs, the one or more programs comprising instructions,
which when executed, cause a server system to: receive, over a
network from a user device remote from the server system, a first
authentication request to authenticate an identity of a user,
wherein the first authentication request includes at least a first
image that corresponds to a first identification document;
determine, using a machine learning system configured to detect one
or more validation faults of a plurality of fault types, a
confidence value that indicates a probability that the first image
includes a first validation fault of the plurality of fault types;
determine whether the confidence value exceeds a threshold
uncertainty value; and in accordance with a determination that the
confidence value exceeds the threshold uncertainty value, transmit,
to a validation system, the first image.
10. The non-transitory computer readable storage medium of claim 9,
wherein the first validation fault of the plurality of fault types
identifies a corresponding fault location.
11. The non-transitory computer readable storage medium of claim 9,
wherein the one or more programs include instructions that cause
the server system to, in accordance with a determination that the
confidence value does not exceed the threshold uncertainty value,
transmit the first authentication request to a remote device.
12. The non-transitory computer readable storage medium of claim 9,
wherein the one or more programs include instructions that cause
the server system to, in accordance with a determination that the
confidence value exceeds the threshold uncertainty value: determine
whether prompting criteria are met; and, in accordance with a
determination that the prompting criteria are met, transmitting
information for displaying a prompt for information that
corresponds to the first validation fault of the plurality of fault
types in the first identification document.
13. The non-transitory computer readable storage medium of claim 9,
wherein the one or more programs include instructions that cause
the server system to: determine, using the machine learning system,
a second validation value that corresponds to a probability that
the first image of the first identification document includes a
second validation fault of the plurality of fault types, distinct
from the first validation fault of the plurality of fault
types.
14. The non-transitory computer readable storage medium of claim
13, wherein the one or more programs include instructions that
cause the server system to: analyze, using the machine learning
system, a set of incoming authentication requests; and determine,
by the machine learning system, a subset of the set of incoming
authentication requests that correspond to invalid authentication
request activity, wherein the second validation fault of the
plurality of fault types is an invalidity characteristic of the
subset of the set of incoming authentication requests.
15. The non-transitory computer readable storage medium of claim 9,
wherein the one or more programs include instructions that cause
the server system to: generate, using the machine learning system,
a set of test authentication requests, wherein each test
authentication request of the set of test authentication requests
includes a respective test validation fault of the plurality of
fault types; analyze, by the machine learning system, the set of
test authentication requests to generate a set of validation values
that corresponds to the set of test authentication requests;
determine that a respective validation value of the set of
validation values does not meet fault review criteria; and in
accordance with the determination that the respective validation
value of the set of validation values does not meet the fault
review criteria, adjusting the machine learning system using the
respective test validation fault of the plurality of fault types
that corresponds to the respective validation value.
16. A server system, comprising: one or more processors; memory;
and one or more programs, wherein the one or more programs are
stored in the memory and are configured for execution by the one or
more processors, the one or more programs including instructions
for: receiving, over a network from a user device remote from the
server system, a first authentication request to authenticate an
identity of a user, wherein the first authentication request
includes at least a first image that corresponds to a first
identification document; determining, using a machine learning
system configured to detect one or more validation faults of a
plurality of fault types, a confidence value that indicates a
probability that the first image includes a first validation fault
of the plurality of fault types; determining whether the confidence
value exceeds a threshold uncertainty value; and in accordance with
a determination that the confidence value exceeds the threshold
uncertainty value, transmitting, to a validation system, the first
image.
17. The server system of claim 16, wherein the first validation
fault of the plurality of fault types identifies a corresponding
fault location
18. The server system of claim 16, wherein the one or more programs
include instructions for, in accordance with a determination that
the confidence value does not exceed the threshold uncertainty
value, transmitting the first authentication request to a remote
device.
19. The server system of claim 16, wherein the one or more programs
including instructions for, in accordance with a determination that
the confidence value exceeds the threshold uncertainty value:
determining whether prompting criteria are met; and, in accordance
with a determination that the prompting criteria are met,
transmitting information for displaying a prompt for information
that corresponds to the first validation fault of the plurality of
fault types in the first identification document.
20. The server system of claim 16, wherein the one or more programs
including instructions for: determining, using the machine learning
system, a second validation value that corresponds to a probability
that the first image of the first identification document includes
a second validation fault of the plurality of fault types, distinct
from the first validation fault of the plurality of fault
types.
21. The server system of claim 20, wherein the one or more programs
including instructions for: analyzing, using the machine learning
system, a set of incoming authentication requests; and determining,
by the machine learning system, a subset of the set of incoming
authentication requests that correspond to invalid authentication
request activity, wherein the second validation fault of the
plurality of fault types is an invalidity characteristic of the
subset of the set of incoming authentication requests.
Description
RELATED APPLICATION
[0001] This application is a continuation of U.S. patent
application Ser. No. 15/993,366 filed May 30, 2018, entitled
"Machine Learning for Document Authentication", which is
incorporated by reference herein in its entirety.
TECHNICAL FIELD
[0002] This application relates generally to user authentication,
and more particularly, to authentication using machine learning to
generate algorithms for analyzing image data.
BACKGROUND
[0003] Identification and verification of remote users is important
for many network communications and transactions. This is
especially true with sensitive communications and important
transactions when users are unacquainted and remote from each
other. Traditionally, a person presents a physical identification
document for inspection by an agent who compares a facial image of
the identification document to the physical appearance of the
person. However, the conventional mechanisms of identifying and
verifying users are inconvenient and inefficient, and result in
burdens for users.
SUMMARY
[0004] Accordingly, there is a need for systems and/or devices with
more efficient, accurate, and intuitive methods for analyzing an
image for authentication of a user. Such systems, devices, and
methods optionally complement or replace conventional systems,
devices, and methods for analyzing an image for authentication of
the user.
[0005] In some embodiments, a machine learning system is used to
generate a model that is used for analyzing image data. The image
data includes, for example, an image of an identification document
(e.g., an identification document, such as a passport or other
government-issued document, that includes an image of a user)
and/or an image captured of the user (e.g., a recent "selfie" image
of the user). A computing system that implements machine learning
analyzes a set of training data that include image data (e.g.,
image data included in a large number of authentication requests).
The machine learning system is used to generate a model that can be
used for analysis of incoming authentication requests. In some
embodiments, a supervised training module of the machine learning
system trains an authentication model using human tagged data
(e.g., by iteratively classifying the data, comparing test data
against human tags, and performing optimization to minimize
classification error). In some embodiments, an unsupervised
training module of the machine learning system trains an
authentication model without using human input for classification
(e.g., by encoding and clustering data to find potential groups and
outliers). In some embodiments, the results of applying an
authentication model (e.g., trained by supervised learning and/or
unsupervised learning) to a set of authentication request data is
compared against the results of human review of the same set of
authentication request data, and discrepancies between the machine
learning results and the human review results are used to alter the
authentication model.
[0006] The disclosed subject matter includes, in one aspect, a
computerized method for receiving first authentication information
for a first transaction. The first authentication information
includes at least a first image that corresponds to a first
identification document. The method additionally includes
receiving, from a validation system, first validation information
that corresponds to a first validation fault. The method
additionally includes storing, by data storage of a machine
learning system, the first validation information. The method
additionally includes receiving second authentication information
for a second transaction. The second authentication information
includes a second image that corresponds to a second identification
document. The method additionally includes determining, using the
machine learning system, a first validation value that corresponds
to a probability that the second image includes the first
validation fault. The method additionally includes determining,
using the first validation value, whether fault review criteria are
met. The method additionally includes, in accordance with a
determination that the fault review criteria are met, transmitting,
to the validation system, the second image.
[0007] In some embodiments, a computer readable storage medium
stores one or more programs. The one or more programs comprise
instructions, which, when executed, cause a device to receive first
authentication information for a first transaction. The first
authentication information includes at least a first image that
corresponds to a first identification document. The one or more
programs additionally comprise instructions that cause the device
to receive, from a validation system, first validation information
that corresponds to a first validation fault. The one or more
programs additionally comprise instructions that cause the device
to store, by data storage of a machine learning system, the first
validation information. The one or more programs additionally
comprise instructions that cause the device to receive second
authentication information for a second transaction. The second
authentication information includes a second image that corresponds
to a second identification document. The one or more programs
additionally comprise instructions that cause the device to
determine, using the machine learning system, a first validation
value that corresponds to a probability that the second image
includes the first validation fault. The one or more programs
additionally comprise instructions that cause the device to
determine, using the first validation value, whether fault review
criteria are met. The one or more programs additionally comprise
instructions that cause the device to, in accordance with a
determination that the fault review criteria are met, transmit, to
the validation system, the second image.
[0008] In some embodiments, a system includes one or more
processors, memory, and one or more programs. The one or more
programs are stored in the memory and configured to be executed by
the one or more processors. The one or more programs include
instructions for receiving first authentication information for a
first transaction. The first authentication information includes at
least a first image that corresponds to a first identification
document. The one or more programs additionally include
instructions for receiving, from a validation system, first
validation information that corresponds to a first validation
fault. The one or more programs additionally include instructions
for storing, by data storage of a machine learning system, the
first validation information. The one or more programs additionally
include instructions for receiving second authentication
information for a second transaction. The second authentication
information includes a second image that corresponds to a second
identification document. The one or more programs additionally
include instructions for determining, using the machine learning
system, a first validation value that corresponds to a probability
that the second image includes the first validation fault. The one
or more programs additionally include instructions for determining,
using the first validation value, whether fault review criteria are
met. The one or more programs additionally include instructions
for, in accordance with a determination that the fault review
criteria are met, transmitting, to the validation system, the
second image.
BRIEF DESCRIPTION OF THE DRAWINGS
[0009] So that the present disclosure can be understood in greater
detail, features of various embodiments are illustrated in the
appended drawings. The appended drawings, however, merely
illustrate pertinent features of the present disclosure and are
therefore not limiting.
[0010] FIG. 1 is a system diagram of an authentication system and
its context, in accordance with some embodiments.
[0011] FIG. 2 is a system diagram of a verification device, in
accordance with some embodiments.
[0012] FIG. 3A illustrates a reference image, in accordance with
some embodiments.
[0013] FIG. 3B illustrates an authentication image, in accordance
with some embodiments.
[0014] FIG. 4 illustrates an example of a reference image that
includes a fault, in accordance with some embodiments.
[0015] FIGS. 5-9 illustrate a validation user interface that
receives fault tagging input, in accordance with some
embodiments.
[0016] FIGS. 10-11 illustrate a validation user interface that is
displayed to a user in accordance with a determination that a
validation value determined for a reference image meets review
criteria, in accordance with some embodiments.
[0017] FIG. 12 is a system diagram of a user device that captures a
reference image and/or an authentication image, in accordance with
some embodiments.
[0018] FIG. 13 illustrates a user device that displays a camera
user interface that includes feedback regarding an image captured
by a camera of the user device.
[0019] FIGS. 14A-14C are a flow diagram illustrating a method for
using machine learning to generate an authentication model for
analyzing image data, in accordance with some embodiments.
[0020] FIG. 15 is a flow diagram illustrating a method for using
machine learning to provide feedback at a device that captures an
image for an authentication request, in accordance with some
embodiments.
[0021] FIG. 16 is a flow diagram illustrating a method for
generating adversarial tests to teach a machine learning system, in
accordance with some embodiments.
[0022] In accordance with common practice, some of the drawings may
not depict all of the components of a given system, method, or
device. Finally, like reference numerals denote like features
throughout the specification and figures.
DETAILED DESCRIPTION
[0023] Numerous details are described herein in order to provide a
thorough understanding of the example embodiments illustrated in
the accompanying drawings. However, some embodiments may be
practiced without many of the specific details, and the scope of
the claims is only limited by those features and aspects
specifically recited in the claims. Furthermore, well-known
processes, components, and materials have not been described in
exhaustive detail so as not to unnecessarily obscure pertinent
aspects of the embodiments described herein.
[0024] The systems and methods described herein pertain to machine
learning algorithms for determining validity of one or more images
that correspond to an authentication request. These systems and
methods improve on prior techniques for determining image validity
of images from an authorization request. In some embodiments, an
authentication request is received from a requesting device (e.g.,
in connection with a payment or other secure transaction). In some
embodiments, the authentication request includes an image of an
identification document, such as a passport (e.g., that is
associated with a user for whom a secure transaction is being
performed). In some embodiments, the authentication request
includes an image of the user (e.g., a recent "selfie" image). In
response to the authentication request, an authentication system
determines validity of the image of the identification document
and/or compares the image of the user with the image of the
identification document to determine whether matching criteria are
met. Information about previously analyzed authentication requests
is used by the authentication system to generate a model that is
used to evaluate incoming authentication requests. In some
embodiments, the model is generated via supervised machine learning
(e.g., the model is generated and/or refined using validation
faults labeled by a human reviewer). In some embodiments, the model
is generated via unsupervised machine learning (e.g., the model is
generated based on an unlabeled set of data). The model is used to
determine whether criteria are met for transmitting an image from
an authentication request to a validation system.
[0025] In some embodiments, the authentication systems described
herein decrease the time required for human review of
identification documents (e.g., by using a model generated by a
machine learning system to analyze an image and provide information
to human reviewers about information generated by the analysis)
and/or reduce the extent of human review used for authenticating
identification documents (e.g., by using the model to determine
whether to bypass human review). Using machine learning as
described herein to reduce the extent of human review and/or to
reduce the time required for human review improves the
authentication device by making the processing of authentication
requests faster and more efficient, with less required human
interaction, which in turn reduces the processing and power used by
an authentication server and/or a validation device.
[0026] Additionally, there is a need to analyze identification
documents to determine whether the identification documents are
authentic, while maintaining compliance of the data set with
applicable security standards. For example, the security of
authentication data is maintained in compliance with the payment
card industry data security standard (PCI DSS). Whereas existing
machine learning systems may rely on labels assigned to a public
set of data, PCI DSS compliance requires limited access to
validation data sets. However, the usability of a model generated
by a machine learning system improves as the size of the available
data set increases. Various embodiments described herein
advantageously perform machine learning using data generated by a
limited set of human validation reviewers. Using a model generated
by a machine learning system to perform image analysis reduces the
amount of time required for human review of authorization requests,
which increases the rate of authorization request processing while
maintaining security of processed data.
[0027] FIG. 1 is a system diagram of an authentication server 100
(also referred to herein as a "machine learning system"), in
accordance with some embodiments. The authentication server 100
typically includes a memory 102, one or more processor(s) 104, a
power supply 106, an input/output (I/O) subsystem 108, and a
communication bus 110 for interconnecting these components.
[0028] The processor(s) 104 execute modules, programs, and/or
instructions stored in the memory 102 and thereby perform
processing operations.
[0029] In some embodiments, the memory 102 stores one or more
programs (e.g., sets of instructions) and/or data structures,
collectively referred to as "modules" herein. In some embodiments,
the memory 102, or the non-transitory computer readable storage
medium of the memory 102 stores the following programs, modules,
and data structures, or a subset or superset thereof: [0030] an
operating system 120; [0031] an authentication request data module
122, which stores information for a plurality of authentication
requests 124 (e.g., a first authentication request 124a, a second
authentication request 124b . . . Nth authentication request 124N),
each of which includes information such as identification
information 126 (e.g., unique identification, user name, user
password, user residential information, user phone number, user
date of birth, and/or user e-mail), a reference image 300a, and/or
an authentication image 350a; and [0032] a machine learning module
126 that uses supervised training module 130, unsupervised training
module 132, and/or adversarial training module 134 to generate
authentication model 136 (e.g., by analyzing validation information
128 corresponding to multiple authentication requests 124).
[0033] The above identified modules (e.g., data structures and/or
programs including sets of instructions) need not be implemented as
separate software programs, procedures, or modules, and thus
various subsets of these modules may be combined or otherwise
re-arranged in various embodiments. In some embodiments, the memory
102 stores a subset of the modules identified above. In some
embodiments, a remote authentication database 152 and/or a local
authentication database 142 store a portion or all of one or more
modules identified above (such as authentication request data 122,
validation information 128, and/or authentication model 136).
Furthermore, the memory 102 may store additional modules not
described above. In some embodiments, the modules stored in the
memory 102, or a non-transitory computer readable storage medium of
the memory 102, provide instructions for implementing respective
operations in the methods described below. In some embodiments,
some or all of these modules may be implemented with specialized
hardware circuits that subsume part or all of the module
functionality. One or more of the above identified elements may be
executed by one or more of the processor(s) 104. In some
embodiments, one or more of the modules described with regard to
the memory 102 is implemented in the memory 202 of a validation
device 201 (FIG. 2) and executed by the processor(s) 204 of the
validation device 201. In some embodiments, one or more of the
modules described with regard to the memory 102 is implemented in
the memory 1202 of a user device 156 (FIG. 12) and executed by the
processor(s) 1204 of the user device 156. For example, machine
learning module 126 is distributed across authentication server
100, validation device(s) 201, validation server 200, and/or user
device 156
[0034] In some embodiments, for each received authentication
request 124, machine learning module 126 stores data for training
authentication model, such as: [0035] Channel via which the
authentication request 124 was received (e.g., mobile, web, and/or
application programming interface (API)) [0036] internet protocol
(IP) address from which the authentication request 124 was received
[0037] camera information (e.g., camera model, device of which the
camera is a component, and/or orientation of camera relative to
device); [0038] transaction origination location (e.g., country,
state, province, and/or city); [0039] type of identification
document captured in reference image 300 (e.g., passport, driver
license, and/or corporate identification badge); [0040]
identification document origination location (e.g., country, state,
province, and/or city); and/or [0041] validation information 128
(e.g., including a verification outcome such as verified, fraud
detected, and/or rejected).
[0042] In some embodiments, generating the authentication model 136
includes generating a regression algorithm for prediction of
continuous variables (e.g., perspective transformation of a
reference image 300 and/or a more complex transformation describing
bending of passport pages).
[0043] In some embodiments, the I/O subsystem 108 communicatively
couples the computing system 100 to one or more devices, such as a
local authentication database 142, a remote authentication database
152, a requesting device 154, a user device 156, validation server
200 (e.g., including one or more server components (e.g., one or
more processor(s) and memory) and/or modules described with regard
to authentication server 100) and/or validation device(s) 201 via a
communications network 150 and/or via a wired and/or wireless
connection. In some embodiments, the communications network 150 is
the Internet.
[0044] The communication bus 110 optionally includes circuitry
(sometimes called a chipset) that interconnects and controls
communications between system components.
[0045] Typically, an authentication system for processing
authentication requests includes a server computer system 100 that
is communicatively connected to one or more validation devices 201
(e.g., via a network 150 and/or an I/O subsystem 108). In some
embodiments, the authentication system receives an authentication
request (e.g., from a user device 156 that captures an image of a
user or from a requesting device 154 that receives an image from
user device 156). For example, the authentication request is a
request to authenticate the identity of a user 124 (e.g., a user
that is a party to a transaction or a user that is requesting
access to a system or physical location). Requesting device 154 is,
for example, a device of a merchant, bank, transaction processor,
computing system or platform, physical access system, or another
user. User device 156 is described in further detail with regard to
FIG. 12 below.
[0046] In some embodiments, an authentication request includes a
reference image 300 (e.g., 300a, as illustrated in FIG. 3A, 300b,
as illustrated in FIG. 4, 300c as illustrated in FIG. 10, or 300d,
as illustrated in FIG. 13). For example, reference image 300 is an
image of an identification document for a user 124). In some
embodiments, an authentication request includes an authentication
image 350 (e.g., 350a, as illustrated in FIG. 3B). For example,
authentication image 350 is an image, series of images, and/or
video of the user 124 captured by a user device 156, such as a
recent "selfie" of the user 124. In some embodiments, an
authentication request includes an authentication image 350 and the
authentication system locates a reference image 300 that
corresponds to the user 124 that provided the authentication image
350.
[0047] In some embodiments, the authentication server 100 causes a
validation device 201 to display all or a part of a reference image
300 and/or all or a part of an authentication image 350 for human
review. In some embodiments, the validation device 201 receives
input that corresponds to a determination of whether authentication
is successful (e.g., based on whether reference image 300 is
sufficiently similar to the authentication image 350). In some
embodiments, validation device 201 transmits validation information
128 that corresponds to a determination of whether authentication
is successful (e.g., indicating that a reference image 300 is a
match for authentication image 350, indicating that reference image
300 is not a match for authentication image 350, and/or indicating
that a fault was detected in reference image 300 and/or
authentication image 250).
[0048] FIG. 2 is a system diagram of a validation device 201 that
displays validation information and/or receives validation input,
in accordance with some embodiments. The validation device 201
typically includes a memory 202, one or more processor(s) 204, a
power supply 206, an input/output (I/O) subsystem 208, and a
communication bus 210 for interconnecting these components. In some
embodiments, validation device 201 is coupled to a validation
server 200 (e.g., that receives data from authentication server
100, transmits data to authentication server 100, and/or manages
validation device(s) 201 (e.g., by transmitting data to validation
device(s) 201 and/or receiving data from validation device(s)
201)). In some embodiments, validation device(s) 201 are managed by
authentication server 100 (e.g., including transmitting data to
authentication server 100 and/or receiving data from authentication
server 100).
[0049] The processor(s) 204 execute modules, programs, and/or
instructions stored in the memory 102 and thereby perform
processing operations.
[0050] In some embodiments, the memory 202 stores one or more
programs (e.g., sets of instructions) and/or data structures,
collectively referred to as "modules" herein. In some embodiments,
the memory 202, or the non-transitory computer readable storage
medium of the memory 202 stores the following programs, modules,
and data structures, or a subset or superset thereof: [0051] an
operating system 220; and [0052] a validation user interface 222,
which displays information (e.g., identification information 126b,
a reference image 300b and/or an authentication image 350b) and/or
includes one or more input controls for receiving validation
input.
[0053] The above identified modules (e.g., data structures, and/or
programs including sets of instructions) need not be implemented as
separate software programs, procedures, or modules, and thus
various subsets of these modules may be combined or otherwise
re-arranged in various embodiments. Furthermore, the memory 202 may
store additional modules not described above. In some embodiments,
the modules stored in the memory 202, or a non-transitory computer
readable storage medium of the memory 202, provide instructions for
implementing respective operations in the methods described below.
In some embodiments, some or all of these modules may be
implemented with specialized hardware circuits that subsume part or
all of the module functionality. One or more of the above
identified elements may be executed by one or more of the
processor(s) 204. In some embodiments, one or more of the modules
described with regard to the memory 202 is implemented in the
memory 102 of a server computing system 100 (FIG. 1) and executed
by the processor(s) 104 of the server computing system 100.
[0054] In some embodiments, the I/O subsystem 208 communicatively
couples the validation device 201 to one or more devices (e.g.,
user input device 212, output device 214, and/or server computer
system 100) via a communications network 150 and/or via a wired
and/or wireless connection. In some embodiments, a user input
device 212 and/or an output device 214 are integrated with
validation device 201 (e.g., as a touchscreen display). In some
embodiments, user input device 212 and/or output device 214 are
peripheral devices communicatively connected to validation device
201. In some embodiments, a user input device 212 includes a
keyboard and/or a pointer device such as a mouse, touchpad, and/or
stylus. In some embodiments, output device 214 includes, for
example, a display and/or a speaker.
[0055] In some embodiments, validation user interface 222, user
input device 212, and/or output device 214 are incorporated into
server computer system 100 (e.g., rather than being implemented in
a separate validation device 201).
[0056] The communication bus 210 optionally includes circuitry
(sometimes called a chipset) that interconnects and controls
communications between system components.
[0057] In some embodiments, validation device 201 (and/or
authentication server 100) displays, by a validation user interface
222 displayed by output device 214, identification information 126,
reference image 300, and/or authentication image 350. In some
embodiments, a reviewer viewing validation user interface 222
provides input to validation device via user input device 212. The
input provided by the reviewer includes, for example, information
regarding the validity of reference image 300 (e.g., the results of
one or more validity checks for reference image 300), information
regarding the validity of authentication image 350 (e.g., the
results of one or more validity checks for authentication image
350), and/or information regarding the similarity of reference
image 300 and authentication image 350 (e.g., the results of one or
more similarity checks for reference image 300 and authentication
image 350).
[0058] FIG. 3A illustrates a reference image 300a, in accordance
with some embodiments. Reference image 300a is, for example, an
image of an identification document 302 that includes a facial
image 304 of a user. For example, reference image 300a is an image
of an identification card, a driver's license, a passport, a
financial instrument (e.g., credit card or debit card), or a
facility access card.
[0059] FIG. 3B illustrates an authentication image 350a that
includes a facial image 352 of the user, in accordance with some
embodiments. For example, authentication image 350a is a "selfie"
captured using user device 156. Authentication image 350a is, e.g.,
a still image or an image extracted from a video. In some
embodiments, the authentication image 350a includes a series of
images or a video (e.g., used for determining that the
authentication image 350 meets liveness requirements).
[0060] FIG. 4 illustrates an example of a reference image 300b that
has been captured of the reference document 302 after the reference
document 302 has been punched (as indicated by the puncture hole
404 that is visible in reference image 300b).
[0061] FIGS. 5-9 illustrate a validation user interface 222 that
receives fault tagging input, in accordance with some embodiments.
In some embodiments, validation user interface 222 is displayed by
output device 214 to a human reviewer (e.g., to facilitate review
of authentication information for an authentication request).
[0062] In some embodiments, validation user interface 222 includes
reference image 300b and/or authentication image 350b (e.g., to
allow a reviewer to compare the images and determine whether the
images match). For example, in FIG. 5, validation user interface
222 includes reference image 300b, authentication image 350b, and a
match control 502 (e.g., receive reviewer input indicating whether
or not reference image 300b in a match for authentication image
350b). In some embodiments, validation user interface 222 includes
a progression control 508 (e.g., for advancing to a subsequent set
of authentication review data, a subsequent interface and/or a
subsequent stage of validation user interface 222). For example, in
a case where no image fault is present in reference image 300b or
authentication image 350b, a human reviewer may use match control
502 to indicate whether or not authentication image 350b is a match
for reference image 300b and then use progression control 508 to
advance to a subsequent authentication review.
[0063] In some embodiments, validation user interface 222 includes
one or more controls for indicating that the human reviewer has
detected a fault and/or for identifying a fault. For example, in
FIG. 5, validation user interface 222 includes an image fault
detection control 504 (e.g., a selectable button) that, when
activated, causes validation device 201 to store and/or transmit
data indicating that a fault has been detected. In some
embodiments, activation of fault detection control 504 initiates a
fault location indication mode. For example, in FIG. 5, because
reference image 300b includes an indication of a puncture hole 404,
the human reviewer has provided input using a user input device 212
(e.g., as indicated by pointer 506) at a location that corresponds
to image fault detection control 504.
[0064] In FIG. 6, in response to detection of the user input at the
location that corresponds to image fault detection control 504, a
fault location indication stage of validation user interface 222
has been displayed. In some embodiments, validation user interface
222 (e.g., image fault detection control 504) and/or pointer 506 is
visually altered to indicate that fault location indication mode is
active. In some embodiments, validation user interface 222 displays
instructions to select the fault location, as indicated at 602.
User input device 212 has been used to select a region that
corresponds to punch mark 404, as indicated by the selection box
604. In some embodiments, multiple fault regions may be selected in
reference image 300b and/or in authentication image 350b.
[0065] In FIG. 7, while the region that corresponds to the image
fault (e.g., punch 404) is selected, input by user input device 212
is detected at a location that corresponds to progression control
508 (e.g., while pointer 506 is at the location that corresponds to
progression control 508).
[0066] In FIG. 8, in response to detection of the input by user
input device 212 at the location that corresponds to progression
control 508, a fault type identification stage of validation user
interface 222 has been displayed. In some embodiments, in the fault
type identification stage of validation user interface 222, a
plurality of fault types (e.g., as discussed further below with
regard to operation 1206) are displayed as selectable options. In
some embodiments, validation user interface 222 displays
instructions to select the fault type, as indicated at 802. In FIG.
8, an input by the human reviewer has selected the "punched
document" fault type, as indicated by the "selected" state of
toggle 804. In some embodiments, multiple fault types may be
assigned to a fault identified in a selected fault region (e.g., as
indicated by selection box 604).
[0067] In FIG. 9, while the fault type that corresponds to the
image fault is selected, input by user input device 212 is detected
at a location that corresponds to progression control 508 (e.g.,
while pointer 506 is. at the location that corresponds to
progression control 508). In some embodiments, in response to
detection of the input at the location that corresponds to
progression control 508, validation information 128 is transmitted
from validation device 201 to authentication server 100 (e.g., for
use by machine learning module 126). For example, the validation
information 128 includes the image 300b, information indicating the
fault type identified as described with regard to FIG. 8, and/or
information indicating the region selected by selection box
604.
[0068] It will be recognized that the description of validation
user interface 222 provided with reference to FIGS. 5-9 is one
illustrative example of an approach to a user interface for image
fault tagging, and that various controls, inputs, and/or sequences
of operations are used in accordance with various embodiments for
human tagging of image faults. For example, in some embodiments, an
indication that an image fault has been detected may be provided by
selection of at least a portion of reference image 300b or at least
a portion of authentication image 350b (e.g., without display
and/or activation of a control 504 for indicating detection of an
image fault).
[0069] In some embodiments, the machine learning module 126
includes a supervised training module 130 that uses image fault
data generated by human reviewers (e.g., as described with regard
to FIG. 5-9) to train authentication model 136 to identify image
faults. For example, the machine learning module 126 uses the
human-tagged image fault data to generate rules of an
authentication model 136 for determining the presence and/or the
location of faults in images included in an authentication request
124. In some embodiments, the machine learning module 126 applies
the generated authentication model 136 to perform image analysis
(e.g., using computer vision techniques) for identifying image
faults in images (such as a reference image 300 and/or an
authentication image 350) of a received authentication request 124.
In some embodiments, in accordance with a determination by machine
learning module 126 that a fault is present in an image, the image
that includes the fault is displayed to a human reviewer for
further review (e.g., with a prompt to identify a location of
fault). In some embodiments, in accordance with a determination by
machine learning module 126 that no fault is present in an image,
the image that includes the fault is displayed to a human reviewer
for further review (e.g., with a prompt to identify a location of
fault). In some embodiments, the human input is fed back into the
supervised machine learning system.
[0070] In some embodiments, the machine learning module 126
analyzes a set of incoming authentication requests 124 (e.g.,
authentication requests received during a predetermined window of
time). In some embodiments, machine learning module 126 includes an
unsupervised training module 132 that trains authentication model
136 using the set of incoming authentication requests without using
human-tagged validation information (e.g., in addition to or in
lieu of a supervised training module 130 that trains authentication
model 136 using human-tagged validation information (e.g., as
described with regard to FIGS. 5-9)).
[0071] In some embodiments, the machine learning module 126
generates multiple authentication models 136. For example, a
respective authentication model 136 is customized based on one or
more parameters (e.g., as defined for an entity (e.g., bank,
merchant, and/or class of users), type of device, type of
identification, and/or type of transaction).
[0072] In some embodiments, the machine learning module 126
determines a subset of the set of incoming authentication requests
124 that correspond to invalid authentication request activity
(e.g., authentication requests that are reported for fraud or that
are determined to include a fault). In some embodiments, the
unsupervised training module 132 determines that, in the subset of
the set of incoming authentication requests 124, a number of
invalid authentication requests exceeds a threshold number of
invalid submissions and/or a rate of fraudulent activity exceeds a
threshold level.
[0073] In some embodiments, a validation fault (e.g., the second
validation fault described with regard to operation 1430, described
with regard to FIG. 14 below) is an invalidity characteristic
(e.g., an indication of an internet protocol (IP) address,
geographical area, device type (e.g., a model that is no longer
available), camera type, document type, transaction type, time at
which request is received, and/or verification outcome) of the
subset of the set of incoming authentication requests. In some
embodiments, weights are applied to the first validation fault, the
second validation fault, and/or any additional validation faults.
In some embodiments, weights are adjusted over time (e.g., in
response to regional activity, transaction types determined to be
problematic, etc.).
[0074] FIGS. 10-11 illustrate a validation user interface 222 that
is displayed to a user in accordance with a determination that a
validation value determined for reference image 300c meets review
criteria, in accordance with some embodiments. For example, machine
learning module 126 has determined that a fault is present in
reference document 1002 captured in image 300c (e.g., reference
document 302 has been punched, as indicated by punch mark 1004). In
some embodiments, validation user interface 222 is displayed by
output device 214 to a human reviewer. For example, in FIG. 10 the
validation user interface 222 is displayed to the human reviewer
with an image fault detected label 1012 and a task prompt 1010
(e.g., prompting the reviewer to confirm presence of the fault
and/or to identify a location of the fault).
[0075] In FIG. 11, user input device 212 has been used (e.g., as
indicated by pointer 1008) to select a region (e.g., as indicated
by the selection box 1006) that corresponds to the image fault
(e.g., punch mark 1004).
[0076] In some embodiments, a machine learning system is used to
analyze images captured by a user device 156 to provide feedback
regarding the captured image at the mobile device. For example, the
feedback is provided prior to generation of an authorization
request.
[0077] FIG. 12 is a system diagram of user device 156 (e.g., a
device that includes and/or is connected to a camera, such as a
personal computer or mobile device) that captures a reference image
300d and/or an authentication image 350d, in accordance with some
embodiments. The user device 156 typically includes a memory 1202,
one or more processor(s) 1204, a power supply 1206, an input/output
(I/O) subsystem 1208, and a communication bus 1210 for
interconnecting these components.
[0078] The processor(s) 1204 execute modules, programs, and/or
instructions stored in the memory 1202 and thereby perform
processing operations.
[0079] In some embodiments, the memory 1202 stores one or more
programs (e.g., sets of instructions) and/or data structures,
collectively referred to as "modules" herein. In some embodiments,
the memory 1202, or the non-transitory computer readable storage
medium of the memory 1202, stores the following programs, modules,
and data structures, or a subset or superset thereof: [0080] an
operating system 1220; [0081] a camera operation module 1224 that
controls camera functioning (e.g., image capture by camera 1214 and
and/or one or more camera parameters of camera 1214); [0082] image
storage module 1226 that stores images captured by camera 1214
(e.g., captured reference image 300d and/or a captured
authentication image 350d); [0083] a machine learning module 1228
that includes an error correction model 1230 for analyzing captured
images to determine presence of image errors; and [0084] a camera
user interface 1232 that displays captured images (e.g., captured
reference image 300d and/or a captured authentication image 350d)
and/or displays feedback to a user regarding the captured
images.
[0085] The above identified modules (e.g., data structures, and/or
programs including sets of instructions) need not be implemented as
separate software programs, procedures, or modules, and thus
various subsets of these modules may be combined or otherwise
re-arranged in various embodiments. Furthermore, the memory 1202
may store additional modules not described above. In some
embodiments, the modules stored in the memory 1202, or a
non-transitory computer readable storage medium of the memory 1202,
provide instructions for implementing respective operations in the
methods described below. In some embodiments, some or all of these
modules may be implemented with specialized hardware circuits that
subsume part or all of the module functionality. One or more of the
above identified elements may be executed by one or more of the
processor(s) 1204.
[0086] In some embodiments, one or more of the modules described
with regard to the memory 1202 (for example, the machine learning
module 1228) is implemented in the memory 102 of authentication
server 100 (FIG. 1) and executed by the processor(s) 104 of the
server computing system 100. For example, authentication server 100
uses machine learning module 1228 and/or machine learning module
126 to analyze a plurality of authorization requests 124 and/or to
generate error correction module 1230. In some embodiments, error
correction model 1230 and/or updates to error correction model 1230
are transmitted from authentication server 100 to user device 156.
In some embodiments, images captured by user device 156, including
images that include errors, are transmitted by user device 156 to
authentication server 100 (e.g., for training error correction
module 1230 and/or included in an authentication request). In some
embodiments, images captured by user device 156 are used by machine
learning module 1228 executing on user device 156 to train error
correction model 1230. In some embodiments, error correction model
1230 is applied by user device 156 to an image captured by user
device 156 (e.g., to provide feedback to a user regarding the
quality of the captured image and/or any faults detected in the
captured image). In some embodiments, error correction model 1230
is applied to a set of images, a human reviewer reviews the same
set of images (e.g., using validation device 200), and error
correction model 1230 is updated based on inconsistencies between
the human review results and error correction model results (e.g.,
difference in confidence level exceeding a threshold inconsistency
value).
[0087] In some embodiments, the I/O subsystem 1208 communicatively
couples the validation device 1202 to one or more devices (e.g.,
user input device 1212, camera 1214, and/or output device 1216) via
a communications network 150 and/or via a wired and/or wireless
connection. In some embodiments, a user input device 1212 and/or an
output device 1216 are integrated with user device 156 (e.g., a
touchscreen display). In some embodiments, user input device 1212
and/or output device 1216 are peripheral devices communicatively
connected to user device 156. In some embodiments, a user input
device 1212 includes a keyboard and/or a pointer device such as a
mouse, touchpad, and/or stylus. In some embodiments, user input
device 1212 includes a microphone. In some embodiments, output
device 1216 includes a display and/or a speaker.
[0088] The communication bus 1210 optionally includes circuitry
(sometimes called a chipset) that interconnects and controls
communications between system components.
[0089] FIG. 13 illustrates a user device 156 that displays a camera
user interface 1232 that includes feedback (e.g., a feedback
message 1304) regarding an image (e.g., reference image 300d)
captured by a camera of user device 156. For example, the displayed
user interface 1302 includes a reference image 300d in which
reference document 1306 is partially covered, as indicated by the
blank space at 1308. In some embodiments, user device 156 applies
an error correction model 1230 to analyze a captured image (e.g.,
after capturing the image). In some embodiments, user device 156
transmits a captured image to authentication server 100,
authentication server 100 uses error correction model 1230 to
analyze the captured image, and authentication server 100 transmits
image error information to user device 156.
[0090] In some embodiments, in accordance with a determination that
one or more errors are identified in a captured image (e.g., one or
more errors indicated in image error information received from
authentication server 100 and/or one or more errors determined by
user device 156 using error correction model 1230), camera
operation module 1224 adjusts one or more parameters of camera 1214
that correspond to at least one of the one or more errors. For
example, in accordance with a determination that a glare error is
identified in a captured image, camera operation module 1224
disables the flash. In some embodiments, camera operation module
1224 adjusts one or more camera parameters and (e.g., subsequently)
automatically captures a new image.
[0091] In some embodiments, in accordance with a determination that
one or more errors are identified in a captured image, camera user
interface 1232 displays a control (e.g., image re-capture button
1310) for capturing a new image. For example, user device 156
displays image re-capture button 1310 that, when activated (e.g.,
by a contact on a touchscreen display of user device 156 that
corresponds to a location of image re-capture button), causes a new
image to be captured by the camera of user device 156 and/or that
causes a camera operation user interface (not shown) to be
displayed.
[0092] FIGS. 14A-14C are a flow diagram illustrating a method 1400
for using machine learning to generate an authentication model 136
for analyzing image data, in accordance with some embodiments. The
method is performed at an authentication server 100, a validation
server 200, user device 156, and/or a validation device 201. For
example, instructions for performing the method are stored in the
memory 102 and executed by the processor(s) 104 of the
authentication server computer system 100. In some embodiments,
part or all of the instructions for performing the method are
stored in the memory 202 and executed by the processor(s) 204 of
the validation device(s) 201. In FIGS. 14A-14C, dotted lines are
used to indicate optional operations.
[0093] The device receives (1402) first authentication information
(e.g., authentication request 124a) for a first transaction. The
first authentication information includes at least a first image
(e.g., a still image, a video, and/or multiple frames from a video)
that corresponds to a first identification document 302. For
example, the first image is a reference image 300, such reference
image 300b described with regard to FIG. 4. In some embodiments,
the authentication request 124 is received from requesting device
154 or user device 156.
[0094] In some embodiments, the device (1404) causes the first
image to be displayed by a display (e.g., output device 214) of the
validation system (e.g., validation server 200 and/or validation
device 201). For example, the device transmits the first image to
the validation device 201 and/or transmits instructions to display
the first image.
[0095] The device receives (1406) from the validation system (e.g.,
validation server 200 and/or validation device 201), first
validation information 128 (e.g., information indicating a type of
fault and/or a location of the fault) that corresponds to a first
validation fault (e.g., a punch mark 404 as described with regard
to FIGS. 4-9). In some embodiments, the first validation
information indicates a location in the first image (e.g.,
reference image 300b) that corresponds to the first validation
fault (e.g., punch mark 404). In some embodiments, the validation
information 128 includes the first image.
[0096] In some embodiments, validation faults include fraud
indications, including, for example: [0097] an indication that
authentication image 350 is a digital copy of reference image 300;
[0098] an indication that image manipulation occurred in the
received reference image 300 and/or authentication image 350;
[0099] an indication that reference image 300 has been cloned (e.g.
at least a (modified) portion of reference image 300 has been used
to create a submitted authentication image 350) and/or an
indication that authentication image 350 has been cloned (e.g., at
least a (modified) portion of authentication image 350 has been
used to create a submitted reference image 300); [0100] an
indication that an image submitted as reference image 300 and/or
authentication image 350 has been spliced (e.g., the image is
generated from at least a portion of an image that is distinct from
reference image 300 and authentication image 350); [0101] an
indication that text has been inserted into (and/or is missing
from) reference image 300 and/or authentication image 350; [0102]
an indication that the reference document depicted in reference
image 350 is a punched document (e.g., a passport or license that
has been punched to indicate that the document is no longer valid);
[0103] an indication that data-masking (blurring, masking, and/or
pixelization) has been applied to an image (e.g., reference image
300 and/or authentication image 350); [0104] an indication that an
irregularity is detected in a feature of reference image 300 that
corresponds to a feature of the document type of reference document
302; [0105] an indication that one or more text features (e.g.,
font type, font size, inclination, and/or kerning) that corresponds
to the reference document 302 is different from a font type
detected in reference image 300; and/or [0106] an indication that a
layout irregularity is detected in reference image 300 (e.g.,
variation from an average layout determined for a document type of
reference document 302 and/or an incorrect location of data and/or
graphical elements of reference document 302).
[0107] In some embodiments, validation faults include readability
faults, including, for example: [0108] a document finding error
(e.g., no document is found in reference image 300 and/or only a
portion of a document is found in reference image 300); [0109] an
optical character recognition (OCR) fault (e.g., text in image not
recognized by OCR and/or no text detected in reference image 300 of
an identification document of a type that requires text); [0110]
blurriness (e.g., document does not meet sharpness and/or contrast
criteria); [0111] motion blur (e.g., smearing or other indication
that document moved during image and/or video capture); [0112]
glare (e.g., a portion of an image and/or text in the image is
obscured by glare), and/or [0113] unintentional data-masking (e.g.,
fingers or other objects are visible in the submitted image, a
portion of the reference document 302 is covered, and/or an
insufficient portion of the reference document 302 or face 352 is
visible in the image).
[0114] The device stores (1408), by data storage of a machine
learning system (e.g., memory 102 of authentication server 100 that
includes machine learning module 126), the first validation
information 128. In some embodiments, the stored first validation
information is used by machine learning module 126 (e.g., using
supervised training module 130) to train authentication model 136.
In some embodiments, the machine learning module 126 uses
validation information 128 that corresponds to a plurality
(typically a very large number) of authentication requests 124 to
generate authentication model 136. For example, data points from
millions of transactions are used to train the authentication model
136.
[0115] The device receives (1410) second authentication information
(e.g., authentication request 124b) for a second transaction. The
second authentication information includes a second image (e.g., as
shown at 300c in FIG. 10) that corresponds to a second
identification document (e.g., distinct from the first
identification document).
[0116] In some embodiments, the machine learning module 126, having
trained authentication module 126 to recognize the punched document
fault based on input received via validation user interface 222
(e.g., as described with regard to FIGS. 5-9), uses the trained
authentication model 136 to determine whether the same fault is
present in an image included in an incoming authentication request
(e.g., as described with regard to FIGS. 10-11).
[0117] The device determines (1412), using the machine learning
system, a first validation value that corresponds to a probability
that the second image includes the first validation fault. In some
embodiments, the machine learning module 126 applies authentication
model 136 trained using supervised training 130. For example, the
machine learning module 126 applies authentication model 136, which
has been trained using human tagged fault information (e.g., fault
information indicating the punch mark 404 in reference image 300b)
included in validation information 128, to determine a confidence
value indicating a probability that the validation fault (e.g.,
punch mark 1004) is present in the second image (e.g., reference
image 300c). In some embodiments, the machine learning module 126
applies authentication model 136 trained using unsupervised
training 132. Typically, validation information (e.g., indicating
presence and/or location of punch marks) indicating information
regarding faults detected in image from a large number of
authentication requests is used to train the authentication model
136 (e.g., prior to using authentication model 136 to identify a
fault in reference images such as reference image 300c). In some
embodiments, the device determines (e.g., simultaneously or
substantially simultaneously), using the machine learning system,
validation values for a set of validation faults that includes the
first validation fault and at least one additional validation fault
(e.g., the second validation fault as described with regard to
1430).
[0118] The device determines (1414), using the first validation
value, whether fault review criteria are met. For example, the
fault review criteria include a criterion that is met in accordance
with a determination that the first validation value exceeds a
threshold uncertainty value. In accordance with a determination
that the fault review criteria are met (e.g., indicating that human
review is required), the method proceeds to operation 1420. In some
embodiments, in accordance with a determination that the fault
review criteria are not met, the method proceeds to operation 1416.
In some embodiments, the device determines whether fault review
criteria are met for each respective validation fault of a set of
validation faults that includes the first validation fault and at
least one additional validation fault.
[0119] In some embodiments, in accordance with a determination that
the fault review criteria are not met, the second image is
displayed (1416) without information indicating that a fault is
detected. For example, authentication device 100 transmits
information to validation device 201 (e.g., information indicating
that fault review criteria are not met and/or that no fault is
detected) that causes validation device 201 to display a validation
user interface 222 as described with regard to FIG. 5. In some
embodiments, validation user interface 222 (e.g., as described with
regard to FIG. 5) is displayed without image fault detection
control 504 (e.g., because no fault is detected). In some
embodiments, validation user interface 222 (e.g., as described with
regard to FIG. 5) is displayed with image fault detection control
504 (e.g., to allow the human reviewer to override the
determination by the machine learning module 126).
[0120] In some embodiments, in accordance with a determination that
the fault review criteria are not met, the device (1418) transmits
authentication information to a remote device. For example,
authentication server 100 transmits authentication information to
requesting device 154 and/or user device 156. In some embodiments,
the authentication information includes information indicating that
no faults are detected in the image. In some embodiments, the
authentication information includes information indicating a
validation value corresponding to a confidence level that no faults
are detected in the image. In some embodiments, the authentication
information includes information indicating that an authentication
request is approved (e.g., in accordance with a determination that
no fault is detected).
[0121] In some embodiments, in accordance with a determination that
the fault review criteria are met, the device transmits (1420), to
the validation system (e.g., for display by the first display or at
a second display that is distinct from the first display), the
second image. In some embodiments, authentication device 100
transmits information indicating that fault review criteria are
met, that one or more faults are detected, and/or determined fault
types of one or more detected faults. In some embodiments, the
information transmitted by authentication device 100 to the
validation system causes validation device 201 to display a
validation user interface 222 as described with regard to FIGS.
10-11. For example, after the device transmits the second image,
validation user interface 222 displays a second image 300c that is
distinct from image 300b that was used to train the authentication
model 136. In some embodiments, in accordance with a determination
that the fault review criteria are met for one or more respective
validation faults of a set of validation faults that includes the
first validation fault and at least one additional validation
fault, information corresponding to the one or more respective
validation faults is transmitted to the validation system. In some
embodiments, in accordance with a determination that the fault
review criteria are met, the device transmits (1422) information
indicating that a fault is detected in the second image. In some
embodiments, the information indicating that the fault is detected
in the second image is used by validation device 201 to display an
indication that a fault is detected, highlighting a location of the
fault, indicating a type of the fault, and or requesting
confirmation of the presence, location, and/or type of the fault.
For example, in FIGS. 10-11, indication 1004 (which indicates that
an image fault has been detected) is displayed. In some
embodiments, the device receives (1424) information from the
validation system (e.g., information provided by a human reviewer
via validation user interface 222) confirming the detected
fault.
[0122] In some embodiments, in accordance with a determination that
the fault review criteria are met, the device determines whether
prompting criteria are met (e.g., based on the validation fault
type and/or based on the validation value). In some embodiments, in
accordance with a determination that prompting criteria are met,
the device transmits (1426) information that causes a fault
identification prompt (e.g., prompt 1006) to be displayed by the
validation device 201.
[0123] In some embodiments, the device receives (1428) information
from the validation system that corresponds to the first validation
fault. For example, information corresponding to a location of the
first validation fault (e.g., as indicated by selection box 1004,
FIG. 11) and/or confirmation of a suggested validation type is
transmitted by validation device 201 to authentication server 100
(e.g., via validation server 200). In some embodiments, the
authentication server stores (e.g., by machine learning module
126), the information (e.g., received from validation device 201)
that corresponds to the first validation fault.
[0124] In some embodiments, the device determines (1430) (e.g.,
using the machine learning module 126) a second validation value
that corresponds to a probability that the second image of the
second identification document includes a second validation fault,
wherein the second validation value is used for determining whether
the fault review criteria are met. For example, machine learning
module 126 trains authentication model 136 to recognize a
validation fault as described with regard to operation 1406 (e.g.,
a validation fault other than the punched document fault, such as a
layout irregularity fault) based on input received via validation
user interface 222 (e.g., as described with regard to FIG. 9). The
machine learning module 126 uses the trained authentication model
136 to determine whether the second fault (e.g., the layout
irregularity fault) is present in an image included in the received
second authentication request.
[0125] In some embodiments, the device transmits (1432) information
regarding the second authentication request (e.g., detection of the
fault by the authentication model 136, confirmation of the fault,
identification of the fault, location of the fault (as identified
by authentication model 136 and/or by a human reviewer via
validation user interface 222) and/or indication of unsuccessful
authentication) to a remote device (e.g., the remote device from
which the second authentication request was received, such as
requesting device 154 or user device 156).
[0126] In some embodiments, the method proceeds from operation 1432
to operation 1402, as indicated at C.
[0127] In some embodiments, a quality review comparison is
performed in which one or more results of the determinations made
using machine learning module 126 (e.g., results including
validation information as described with regard to operation 1406
and/or a first validation value as described with regard to
operation 1412) are compared with results of corresponding
determinations made by human reviewers. For example, for a set of
authentication requests evaluated using authentication model 136, a
matching set of authentication requests are provided to human
reviewers (e.g., via a validation user interface 222 as described
with regard to FIGS. 5-9). In some embodiments, discrepancies
between the determinations made using machine learning module 126
and the determinations made by the human reviewers are used to
train the authentication model 136. In some embodiments,
discrepancies between the determinations made using machine
learning module 126 and the determinations made by the human
reviewers are reviewed by a quality control reviewer. In some
embodiments, the quality control review results are used to train
the authentication model 136. For example, in accordance with a
quality control review determination that a result of a review
performed by machine learning module 126 differs from a result
performed by a human reviewer, if the quality control reviewer
determines that the human review result was the correct result, a
weight applied to one or more parameters of authentication model
136 are adjusted.
[0128] It should be understood that the particular order in which
the operations in FIGS. 14A-14C have been described is merely an
example and is not intended to indicate that the described order is
the only order in which the operations could be performed. One of
ordinary skill in the art would recognize various ways to reorder
the operations described herein.
[0129] It will be understood that the flow described with regard to
FIGS. 14A-14C is an example of flow using an exemplary first
transaction and exemplary second transaction. In some embodiments,
authentication server 100, validation server 200, and/or validation
device(s) 201 process many transactions simultaneously.
[0130] FIG. 15 is a flow diagram illustrating a method 1500 for
using machine learning to provide feedback at a device that
captures an image for an authentication request, in accordance with
some embodiments. The method is performed at user device 156,
authentication server 100, validation device(s) 201, and/or
validation server 200. For example, instructions for performing the
method are stored in the memory 1202 and executed by the
processor(s) 1204 of the user device 156. In some embodiments, part
or all of the instructions for performing the method are stored in
the memory 102 and executed by the processor(s) 104 of the
authentication server 100. In FIG. 15, dotted lines are used to
indicate optional operations.
[0131] It should be understood that the particular order in which
the operations in FIG. 15 have been described is merely an example
and is not intended to indicate that the described order is the
only order in which the operations could be performed. One of
ordinary skill in the art would recognize various ways to reorder
the operations described herein.
[0132] In some embodiments, the device captures (1502) an image
and/or video using camera 1214. For example, the device captures a
reference image 300a of an identification document 302, or the
device captures an authentication image 350a (e.g., a "selfie"
image of the user). In some embodiments, the device captures sound
data in addition to image data. For example, the device captures a
video recording of the user that includes an audio recording.
[0133] In some embodiments, the device applies (1504) error
correction model 1230 to the captured image to determine whether
the image includes an image capture error (e.g., fraud indications
and/or readability faults as described with regard to operation
1406).
[0134] In some embodiments, the device determines (1506) (e.g.,
based on the application of the error correction model 1230 to the
captured image) whether the captured image meets quality
requirements. In accordance with a determination that the captured
image does not meet quality requirements (e.g., a validation value
determined using error correction model 1230 for the captured image
is above a threshold uncertainty threshold), the method proceeds to
operation 1508. In accordance with a determination that the
captured image meets quality requirements, the method proceeds to
operation 1516.
[0135] In some embodiments, in accordance with a determination that
the captured image does not meet quality requirements, the device
provides (1508) image capture feedback. In some embodiments, image
capture feedback includes an indication of a detected image capture
error, instructions for avoiding a detected image capture error,
and/or instructions for improving image quality. In some
embodiments, image capture feedback is output by output device 1216
(e.g., a display and/or a speaker). For example, image capture
feedback is displayed in a camera user interface 1232 (e.g., as
illustrated by feedback message 1304) and/or provided via a
text-to-speech module as audio output.
[0136] In some embodiments, in accordance with a determination that
the captured image does not meet quality requirements, the device
adjusts (1510) one or image capture parameters of camera 1214. In
some embodiments, camera operation module 1224 adjusts one or more
parameters (e.g., exposure, white balance, focal length, aperture,
sensitivity (ISO), flash setting, and/or capture mode) of camera
1214. In some embodiments, camera operation module 1224 adjusts one
or more parameters that correspond to at least one of one or more
errors detected using error correction model 1230.
[0137] In an illustrative example, the error correction module 1230
determines, based on training data, that a captured image includes
excessive glare (e.g., because a portion of a facial image is not
visible where facial features would typically be present within a
facial image and/or within a reference document; because a portion
of text is missing where text would typically be present within a
reference document; and/or because a level of contrast in a first
portion of the document is lower than a level of contrast in a
second portion of the document). In accordance with a determination
that the captured image includes excessive glare, camera operation
module 1224 adjusts a property of a flash system (e.g., by
disabling a flash device, reducing flash intensity, activating an
alternative flash mode, and/or activating a different flash
producing device).
[0138] In some embodiments, in accordance with a determination that
the captured image does not meet quality requirements, the device
uses (1512) training information (e.g., the captured image and/or
information regarding a detected image capture error) to train
error correction model 1230. For example, user device 156 stores
the training information or transmits the training information to
authentication sever 100.
[0139] In some embodiments, in accordance with a determination that
the captured image meets quality requirements, the device uses
(1514) quality information (e.g., the captured image and, in some
embodiments, a quality indication that the captured image meets the
quality requirements) to train error correction model 1230. For
example, user device 156 stores the quality information and/or
transmits the quality information to authentication sever 100.
[0140] In some embodiments, the device determines (1516) whether
additional image(s) are needed. For example, if a previously
captured image was a reference image 300 and (no authentication
image 350 has been captured), the device displays a prompt (e.g.,
in camera user interface 1232) to capture the authentication image
350, and if a previously captured image was an authentication image
350 and (no reference image 300 has been captured or previously
provided), the device displays a prompt (e.g., in camera user
interface 1232) to capture the reference image 300. In accordance
with a determination that additional image(s) are needed, the
method proceeds to operation 1502. In accordance with a
determination that no additional image(s) are needed, the method
proceeds to operation 1518.
[0141] In some embodiments, the device generates (1518) an
authentication request 124 using the captured reference image 300
and or authentication image 350 (e.g., the images that met the
quality requirements).
[0142] In some embodiments, the device transmits (1520) the
captured reference image 300 and or authentication image 350 (e.g.,
the images that met the quality requirements) to the requesting
device 154 (e.g., a device that generates an authentication request
124). In some embodiments, the device transmits device information
(e.g., a serial number, model, and/or manufacturer of the device)
and/or camera information (e.g., camera type, camera position
relative to the device, and/or one or more image capture
parameters) to the authentication server 100. In some embodiments,
the device transmits captured audio and/or video to the
authentication server 100. For example, the captured audio and/or
video are used by device 156 and/or by authentication server 100 to
evaluate whether live capture criteria are met.
[0143] FIG. 16 is a flow diagram illustrating a method 1600 for
generating adversarial tests to teach a machine learning system, in
accordance with some embodiments. The method is performed at an
authentication server 100. For example, instructions for performing
the method are stored in the memory 102 and executed by the
processor(s) 104 of the authentication server computer system 100.
In some embodiments, the method 1700 is performed by the machine
learning module 126 (e.g., an adversarial training module 134 of
the machine learning module 126).
[0144] In some embodiments, the device generates (1602) a set of
test authentication requests. Each test authentication request of
the set of test authentication requests includes a respective test
validation fault (e.g., a fault as described with regard to
operation 1406).
[0145] In some embodiments, the device analyzes (1604) the set of
test authentication requests (e.g., by applying authentication
model 136 to one or more images included in the authentication
requests) to generate a set of validation values that corresponds
to the set of test authentication requests. For example, machine
learning module 126 determines validation values for the
authentication requests (e.g., as described with regard to
operation 1412).
[0146] In some embodiments, the device determines (1606) whether a
respective validation value of the set of validation values does
not meet fault review criteria (e.g., criteria that are met in
accordance with a determination that the validation value is not
within a predefined range of a target uncertainty value that
corresponds to a test authentication request). In accordance with
the determination that the respective validation value of the set
of validation values does not meet the fault review criteria, the
device adjusts the machine learning system (e.g., by adjusting
authentication model 136). For example, authentication model 136
includes weights applied to each fault in a set of faults (e.g., as
described with regard to operation 1406), and, in accordance with a
determination that the respective validation value of the set of
validation values does not meet the fault review criteria, the
device adjusts a weight applied to the tested validation fault
and/or adds a new validation fault.
[0147] In an illustrative example, the set of test authentication
requests includes authentication requests in which a text feature
(e.g., font size) in a reference image 300 is altered by varying
degrees from the same text feature as found in a legitimate
identification document. If an invalid text field passes through
the machine learning system without triggering review, the machine
learning system is adjusted to identify the invalid text.
[0148] It should be understood that the particular order in which
the operations in FIG. 16 have been described is merely an example
and is not intended to indicate that the described order is the
only order in which the operations could be performed. One of
ordinary skill in the art would recognize various ways to reorder
the operations described herein.
[0149] Features of the present invention can be implemented in,
using, or with the assistance of a computer program product, such
as a storage medium (media) or computer readable storage medium
(media) having instructions stored thereon/in which can be used to
program a processing system to perform any of the features
presented herein. The storage medium (e.g., the memory 102 and the
memory 202) can include, but is not limited to, high-speed random
access memory, such as DRAM, SRAM, DDR RAM or other random access
solid state memory devices, and may include non-volatile memory,
such as one or more magnetic disk storage devices, optical disk
storage devices, flash memory devices, or other non-volatile solid
state storage devices. In some embodiments, the memory 102 and the
memory 202 include one or more storage devices remotely located
from the CPU(s) 104 and 204. The memory 102 and the memory 202, or
alternatively the non-volatile memory device(s) within these
memories, comprises a non-transitory computer readable storage
medium.
[0150] Communication systems as referred to herein (e.g., the
communication system 108 and the communication system 208)
optionally communicate via wired and/or wireless communication
connections. Communication systems optionally communicate with
networks (e.g., the networks 150 and 152), such as the Internet,
also referred to as the World Wide Web (WWW), an intranet and/or a
wireless network, such as a cellular telephone network, a wireless
local area network (LAN) and/or a metropolitan area network (MAN),
and other devices by wireless communication. Wireless communication
connections optionally use any of a plurality of communications
standards, protocols and technologies, including but not limited to
Global System for Mobile Communications (GSM), Enhanced Data GSM
Environment (EDGE), high-speed downlink packet access (HSDPA),
high-speed uplink packet access (HSUPA), Evolution, Data-Only
(EV-DO), HSPA, HSPA+, Dual-Cell HSPA (DC-HSPDA), long term
evolution (LTE), near field communication (NFC), wideband code
division multiple access (W-CDMA), code division multiple access
(CDMA), time division multiple access (TDMA), Bluetooth, Wireless
Fidelity (Wi-Fi) (e.g., IEEE 102.11a, IEEE 102.11ac, IEEE 102.11ax,
IEEE 102.11b, IEEE 102.11g and/or IEEE 102.11n), voice over
Internet Protocol (VoIP), Wi-MAX, a protocol for e-mail (e.g.,
Internet message access protocol (IMAP) and/or post office protocol
(POP)), instant messaging (e.g., extensible messaging and presence
protocol (XMPP), Session Initiation Protocol for Instant Messaging
and Presence Leveraging Extensions (SIMPLE), Instant Messaging and
Presence Service (IMPS)), and/or Short Message Service (SMS), or
any other suitable communication protocol, including communication
protocols not yet developed as of the filing date of this
document.
[0151] It will be understood that, although the terms "first,"
"second," etc. may be used herein to describe various elements,
these elements should not be limited by these terms. These terms
are only used to distinguish one element from another.
[0152] The terminology used herein is for the purpose of describing
particular embodiments only and is not intended to be limiting of
the claims. As used in the description of the embodiments and the
appended claims, the singular forms "a," "an" and "the" are
intended to include the plural forms as well, unless the context
clearly indicates otherwise. It will also be understood that the
term "and/or" as used herein refers to and encompasses any and all
possible combinations of one or more of the associated listed
items. It will be further understood that the terms "comprises"
and/or "comprising," when used in this specification, specify the
presence of stated features, steps, operations, elements, and/or
components, but do not preclude the presence or addition of one or
more other features, steps, operations, elements, components,
and/or groups thereof.
[0153] As used herein, the term "if" may be construed to mean
"when" or "upon" or "in response to determining" or "in accordance
with a determination" or "in response to detecting," that a stated
condition precedent is true, depending on the context. Similarly,
the phrase "if it is determined [that a stated condition precedent
is true]" or "if [a stated condition precedent is true]" or "when
[a stated condition precedent is true]" may be construed to mean
"upon determining" or "in response to determining" or "in
accordance with a determination" or "upon detecting" or "in
response to detecting" that the stated condition precedent is true,
depending on the context.
[0154] The foregoing description, for purpose of explanation, has
been described with reference to specific embodiments. However, the
illustrative discussions above are not intended to be exhaustive or
to limit the claims to the precise forms disclosed. Many
modifications and variations are possible in view of the above
teachings. The embodiments were chosen and described in order to
best explain principles of operation and practical applications, to
thereby enable others skilled in the art.
* * * * *