Granted Patent B2

US 12,387,515 · App. 17/809,920 · Granted Aug 12, 2025

Inventor: Lawrence Douglas (McLean, VA)

Assignee: Capital One Services, LLC

G06V30/413G06T7/0002G06V20/95G06V30/418G06T2207/20081

Abstract

Claims (96)

1. A system for providing multi-tier machine learning model processing for document authenticity, the system comprising:

one or more processors; and

a non-transitory, computer-readable storage medium storing instructions, which when executed by the one or more processors cause the one or more processors to perform operations comprising:

receiving an image representing an instance of a document,

wherein the document is associated with a plurality of versions of the document,

wherein the plurality of versions of the document include:

a first version associated with a first plurality of security features for authenticating the document, and

a second version associated with a second plurality of security features for authenticating, and

wherein the second plurality of security features are different from the first plurality of security features;

inputting the image into a machine learning model that is trained to (1) determine, using a first tier of the machine learning model, a subset of the first plurality of security features which the instance of the document includes, (2) adjust parameters of a second tier of the machine learning model based on the subset of the first plurality of security features, and (3) determine, using the second tier of the machine learning model, whether the received image represents an authentic document;

receiving, from the machine learning model and based on inputting the image, a probability that the image represents an authentic instance of the document;

determining whether the probability that the image represents the authentic instance of the document meets a threshold;

based on determining that the probability that the image represents the authentic instance of the document meets the threshold, providing, to a user, an indication that the instance of the document is authentic; and

based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold, providing, to the user, indications of one or more security features of the first plurality of security features associated with the first version of the document and instructions for authenticating the instance of the document using the one or more security features.

2. The system of claim 1 , wherein the instructions further cause the one or more processors to perform operations comprising:

based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold:

extracting, from the image, an identifier associated with the instance of the document;

comparing the identifier with a plurality of identifiers corresponding to a plurality of images associated with a plurality of instances of the document;

determining, based on the comparing, that the identifier matches a threshold number of instances of the document; and

based on determining that the identifier matches the threshold number of instances of the document, determining that the instance of the document is not authentic.

3. The system of claim 1 , wherein the instructions further cause the one or more processors to perform operations comprising:

based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold:

extracting, from the image, an identifier associated with the instance of the document;

determining that the identifier matches a stored image of another instance of the document stored in a database, wherein the stored image is associated with a user location and a user time; and

based on determining that the user location or the user time is not within a time location threshold of a current time and a current location, determining that the instance of the document is not authentic.

4. The system of claim 1 , wherein the instructions, when executed by the one or more processors, further cause the one or more processors to perform operations comprising:

receiving, from the machine learning model, one or more indications of one or more areas on the image that contributed to lowering the probability that the instance of the document is authentic; and

generating for display the image of the document with the one or more areas highlighted to the user.

5. A method for providing multi-tier machine learning model processing for document authenticity, the method comprising:

receiving an image representing an instance of a document,

wherein the document is associated with a plurality of versions of the document,

wherein the plurality of versions of the document include:

a first version associated with a first plurality of security features for authenticating the document, and

a second version associated with a second plurality of security features for authenticating, and

wherein the second plurality of security features are different from the first plurality of security features;

inputting the image into a machine learning model,

wherein the machine learning model is trained to (1) determine, using a first tier of the machine learning model, a subset of the first plurality of security features which the instance of the document includes, (2) adjust parameters of a second tier of the machine learning model based on the subset of the first plurality of security features, and (3) determine, using the second tier of the machine learning model, whether the received image represents an authentic document;

receiving, from the machine learning model and based on inputting the image, a probability that the image represents an authentic instance of the document;

determining whether the probability that the image represents the authentic instance of the document meets a threshold; and

based on determining that the probability that the image represents the authentic instance of the document meets the threshold, providing, to a user, an indication that the instance of the document is authentic.

6. The method of claim 5 , further comprising, based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold, providing, to the user, indications of one or more security features of the first plurality of security features associated with the first version of the document and instructions for authenticating the instance of the document using the one or more security features.

7. The method of claim 6 , further comprising:

receiving an input from the user that the document is authentic; and

based on receiving the input, initiating a training routine of the machine learning model, wherein the initiating comprises inputting the image into the machine learning model with a label indicating that the document is authentic.

8. The method of claim 5 , wherein the machine learning model is configured to perform operations comprising:

identifying that the first version is associated with the instance of the document; and

based on the first plurality of security features being associated with the first version, determining the probability that the instance of the document is authentic.

9. The method of claim 5 , further comprising:

based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold:

accessing an identifier associated with the instance of the document;

comparing the identifier with a plurality of identifiers corresponding to a plurality of images associated with a plurality of instances of the document;

determining, based on the comparing, that the identifier matches a threshold number of instances of the document; and

based on determining that the identifier matches the threshold number of instances of the document, determining that the instance of the document is not authentic.

10. The method of claim 5 , further comprising:

based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold:

generating, from the image, a token representing the instance of the document;

determining that the token matches a stored token representing another instance of the document, wherein the stored token is associated with a detection location and a detection time; and

based on determining that the detection location is not within a location threshold of a current location or the detection time is not within a time threshold of a current time, determining that the instance of the document is not authentic.

11. The method of claim 10 , further comprising updating a record associated with the token to indicate a detected location and detected time associated with detection of the token.

12. The method of claim 5 , further comprising:

receiving, from the machine learning model, one or more indications of one or more areas on the image that contributed to lowering the probability that the instance of the document is authentic; and

generating for display the image of the document with the one or more areas highlighted to the user.

13. A non-transitory, computer-readable medium for providing multi-tier machine learning model processing for document authenticity, storing instructions that, when executed by one or more processors, cause the one or more processors to perform operations comprising:

receiving an image representing an instance of a document,

wherein the document is associated with a plurality of versions of the document,

wherein the plurality of versions of the document include:

a first version associated with a first plurality of security features for authenticating the document, and

a second version associated with a second plurality of security features for authenticating, and

wherein the second plurality of security features are different from the first plurality of security features;

inputting the image into a machine learning model,

wherein the machine learning model is trained to (1) determine, using a first tier of the machine learning model, a subset of the first plurality of security features which the instance of the document includes, (2) adjust parameters of a second tier of the machine learning model based on the subset of the first plurality of security features, and (3) determine, using the second tier of the machine learning model, whether the received image represents an authentic document;

receiving, from the machine learning model and based on inputting the image, a probability that the image represents an authentic instance of the document;

determining whether the probability that the image represents the authentic instance of the document meets a threshold; and

based on determining that the probability that the image represents the authentic instance of the document meets the threshold, providing, to a user, an indication that the instance of the document is authentic.

14. The non-transitory, computer-readable medium of claim 13 , wherein the instructions further cause the one or more processors to perform operations comprising, based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold, providing, to the user, indications of one or more security features of the first plurality of security features associated with the first version of the document and instructions for authenticating the instance of the document using the one or more security features.

15. The non-transitory, computer-readable medium of claim 14 , wherein the instructions further cause the one or more processors to perform operations comprising:

receiving an input from the user that the document is authentic; and

based on receiving the input, initiating a training routine of the machine learning model, wherein the initiating comprises inputting the image into the machine learning model with a label indicating that the document is authentic.

16. The non-transitory, computer-readable medium of claim 13 , wherein the machine learning model is configured to perform operations comprising:

identifying that the first version is associated with the instance of the document; and

based on the first plurality of security features being associated with the first version, determining the probability that the instance of the document is authentic.

17. The non-transitory, computer-readable medium of claim 13 , wherein the instructions further cause the one or more processors to perform operations comprising:

based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold:

accessing an identifier associated with the instance of the document;

comparing the identifier with a plurality of identifiers corresponding to a plurality of images associated with a plurality of instances of the document;

determining, based on the comparing, that the identifier matches a threshold number of instances of the document; and

based on determining that the identifier matches the threshold number of instances of the document, determining that the instance of the document is not authentic.

18. The non-transitory, computer-readable medium of claim 13 , wherein the instructions further cause the one or more processors to perform operations comprising:

based on determining that the probability that the image represents the authentic instance of the document does not meet the threshold:

generating, from the image, a token representing the instance of the document;

determining that the token matches a stored token representing another instance of the document, wherein the stored token is associated with a detection location and a detection time; and

based on determining that the detection location is not within a location threshold of a current location or the detection time is not within a time threshold of a current time, determining that the instance of the document is not authentic.

19. The non-transitory, computer-readable medium of claim 18 , wherein the instructions further cause the one or more processors to perform operations comprising updating a record associated with the token to indicate a detected location and detected time associated with detection of the token.

20. The non-transitory, computer-readable medium of claim 13 , wherein the instructions further cause the one or more processors to perform operations comprising:

receiving, from the machine learning model, one or more indications of one or more areas on the image that contributed to lowering the probability that the instance of the document is authentic; and

generating for display the image of the document with the one or more areas highlighted to the user.

Assignments (1)

Continuity (1)

References Cited (13)