IP Library Granted Patent US 12,738,083
Granted Patent B1
US 12,738,083 · App. 18/499,057 · Granted Sep 15, 2026

System and method of knowledge graph based ID anomaly detection in order to determine if the anomaly is associated with ID fraud

Inventors: Ashwin Pingali (Parker, CO); Arman Mohseni-Kabir (Carlsbad, CA); Marzieh Mehdizadeh (Sunnyvale, CA); Lulu Li (Millbrae, CA); Sudha Vijayakumar (Campbell, CA); Osvaldo Driollet (Carlsbad, CA)
Assignee: Jumio Corporation
G06V30/1914G06N5/02
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,738,083
App. No.
18/499,057
Granted
Sep 15, 2026
Kind
B1
Abstract

The disclosure includes a system and method for anomaly detection of an ID using a knowledge graph. Human curated knowledge may be used as a source of information to aid the system. Image fragments are created from an image of an ID. A knowledge graph schema may be used to define node names and relationships. The image fragments are processed and evaluated according to a workflow. The evaluation results can be compared with a gold standard for a good image. The use of a knowledge graph permits patterns in the images to be used to detect anomalies, which may include potential fraud.

Claims (42)

1 . A method of performing identification document (ID) anomaly detection using a knowledge graph, comprising:

generating a hierarchical multidimensional view of identification documents based at least in part on human curated expert knowledge; and

providing a knowledge graph base having a knowledge graph schema to define, for each ID type of a plurality of ID types, node labels for image fragments, features of image fragments, relationships between nodes, and select an execution workflow for creating fragments of images, assigning fragment processing functions to the image fragments, and assigning evaluation functions for processed image fragments;

receiving an image of an ID and a class label for the ID associated with an ID type;

initiating a workflow for a received ID, the workflow selected based on a least one attribute of the ID;

performing a fragment creation process to break down the image of the received ID under test into a set of image fragments according to the workflow;

processing features of the image fragments according to fragment processing functions assigned to the image fragments according to the workflow;

evaluating results of the processed image fragments based on the evaluation functions assigned to the processed image fragments according to the workflow; and

reporting whether the ID has an anomaly associated with ID fraud, wherein a decision includes comparing the evaluation functions against a gold standard of a valid image.

2 . The method of claim 1 , wherein the reporting comprises reporting at least one of a class anomaly and an image anomaly.

3 . The method of claim 2 , comprising creating fingerprints of one or more sub-graphs of the knowledge graph, converting the fingerprints into vectors, comparing the vectors for similarity, and then identifying the sub-graphs for discrepancies and evaluating a risk associated with a discrepancy.

4 . The method of claim 3 , wherein invariant image fragment features comprise spatial features with two-dimensional spatial relationships between the spatial features of different image fragments.

5 . The method of claim 3 , wherein invariant image fragment features comprising color features with two-dimensional relationships between the color features of different image fragments.

6 . The method of claim 1 , wherein the fragment creation process creates a fragment and populates a schema for the fragment with at least one feature property.

7 . The method of claim 1 , where the fragment processing function includes at least one of: text extraction and image processing.

8 . The method of claim 1 , further comprising utilizing graph analytics to learn relationship between fragments of individual IDs and across IDs of a same ID type.

9 . The method of claim 8 , further comprising inferring a set of rules based on learned relationships between fragments.

10 . The method of claim 9 , further comprising generating a reasoning engine to identify deviations in invariant relationships, identifying potential fraudulent manipulations, and identifying quality issues.

11 . The method of claim 1 , further comprising separating an image from a background prior to generating image fragments.

12 . The method of claim 1 , further comprising generating, in test mode, synthetic anomalies and inserting the synthetic anomalies into ID images and fragments.

13 . A system comprising:

a processor; and

a memory, the memory storing instructions that, when executed by the processor, cause the system to:

generate a hierarchical multidimensional view of identification documents based at least in part on human curated expert knowledge; and

utilize a knowledge graph base having a knowledge graph schema to define, for each ID type of a plurality of ID types, node labels for image fragments, features of image fragments, relationships between nodes, select an execution workflow for creating fragments of images, implement a fragment processing function, and implement evaluation functions for processed fragments;

receive an image of an ID;

initiate a workflow for a received ID, the workflow selected based on a least one attribute of the ID;

perform a fragment creation process to break down an image of the received ID under test into a set of image fragments;

process features of the image fragments according to its associated fragment processing functions;

evaluate results of processed image fragments based on an associated evaluation functions; and

decide whether the ID has an anomaly associated with ID fraud, wherein the decision includes comparing the evaluation functions against a gold standard of a valid image.

14 . The system of claim 13 , wherein the image fragments have invariant features.

15 . The system of claim 13 , comprising creating fingerprints of one or more sub-graphs of the knowledge graph base, converting the fingerprints into vectors, comparing the vectors for similarity, and then identifying the sub-graphs for discrepancies and evaluating a risk associated with the discrepancies.

16 . The system of claim 15 , wherein the image fragment features comprise spatial features with two-dimensional spatial relationships between the spatial features of different image fragments.

17 . The system of claim 15 , wherein the image fragment features comprise color features with two-dimensional relationships between the color features of different image fragments.

18 . The system of claim 13 , wherein a fragment creation function creates a fragment and populates the fragment with at least one feature property.

19 . The system of claim 13 , where the fragment processing function includes at least one of text extraction and image processing.

20 . The system of claim 13 , further comprising a graph analytics engine configured to learn relationship between fragments of individual IDs and across IDs of a same type.

21 . The system of claim 20 , wherein the graph analytics engine generates a set of rules based on the learned relationships between fragments.

22 . The system of claim 21 , further comprising a reasoning engine to identify deviations in invariant relationships, identify potential fraudulent manipulations, and identify quality issues.

23 . The system of claim 13 , wherein the image is separated from its background prior to generating image fragments.

24 . The system of claim 13 , wherein the system is configured to have a test mode to generate and insert synthetic anomalies into ID images and fragments.

Assignments (2)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 22, 2023
From: PINGALI, ASHWIN; MOHSENI-KABIR, ARMAN; MEHDIZADEH, MARZIEH; LI, LULU; VIJAYAKUMAR, SUDHA; DRIOLLET, OSVALDO
To: JUMIO CORPORATION
Reel/Frame 065643/0806 →
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Nov 21, 2023
From: PINGALI, ASHWIN; MOHSENI-KABIR, ARMAN; MEHDIZADEH, MARZIEH; LI, LULU; VIJAYAKUMAR, SUDHA; DRIOLLET, OSVALDO
To: JUMIO CORPORATION
Reel/Frame 065637/0929 →
References Cited (34)
US 7769704B2 · Marfatia · 2010 [cited by applicant]
US 10991053B2 · Pingali · 2021 [cited by applicant]
US 11238115B1 · Newman · 2022 [cited by applicant]
US 11816596B2 · Pingali · 2023 [cited by applicant]
US 12112519B1 · Polichroniadis · 2024 [cited by applicant]
US 12147647B2 · Marchetti · 2024 [cited by applicant]
US 20100049538A1 · Frazer · 2010 [cited by applicant]
US 20120158633A1 · Eder · 2012 [cited by applicant]
US 20150317449A1 · Eder · 2015 [cited by applicant]
US 20160321661A1 · Hammond · 2016 [cited by applicant]
US 20160364794A1 · Chari · 2016 [cited by applicant]
US 20190122111A1 · Min · 2019 [cited by applicant]
US 20190259033A1 · Reddy · 2019 [cited by applicant]
US 20190312869A1 · Han · 2019 [cited by applicant]
US 20190340614A1 · Hanis · 2019 [cited by applicant]
US 20190340615A1 · Hanis · 2019 [cited by applicant]
US 20190372940A1 · McDougall · 2019 [cited by applicant]
US 20200401835A1 · Zhao · 2020 [cited by applicant]
US 20210248268A1 · Ardhanari · 2021 [cited by applicant]
US 20210304021A1 · Puri · 2021 [cited by applicant]
US 20220075948A1 · Yuan · 2022 [cited by examiner]
US 20220111960A1 · Tran · 2022 [cited by applicant]
US 20220292262A1 · Japa · 2022 [cited by applicant]
US 20230132720A1 · Khmaissia · 2023 [cited by examiner]
US 20230291756A1 · Monnig · 2023 [cited by applicant]
US 20240013220A1 · Martins · 2024 [cited by applicant]
US 20240221411A1 · Wells · 2024 [cited by examiner]
US 20240303662A1 · Shah · 2024 [cited by applicant]
US 20250200630A1 · Wang · 2025 [cited by applicant]
“Bilot et al., ““Graph Neural Networks for Intrusion Detection: A Survey,”” May 12, 2023, IEEEAccess, vol. 11,2023, pp. 49114-49121.” [cited by applicant]
Li et al., “Recognizing Object by Components With Human Prior Knowledge Enhances Adversarial Robustness of Deep Neural Networks,” Jan. 18, 2023, IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 45, N… [cited by applicant]
Liu et al., “Combining Graph Neural Networks With Expert Knowledge for Smart Contract Vulnerability Detection,” Jul. 7, 2021, IEEE Transactions on Knowledge and Data Engineering, vol. 35, No. 2, Feb. 2023, pp. 1296-1302. [cited by applicant]
Wang et al., “Fake News Detection via Knowledge-driven Multimodal Graph Convolutional Networks,” Jun. 8, 2020, ICMR '20, Oct. 26-29, 2020, Dublin, Ireland, pp. 540-546. [cited by applicant]
Xu et al., “Evidence-aware Fake News Detection with Graph Neural Networks,” Apr. 25, 2022, WWW '22: Proceedings of the ACM Web Conference 2022, pp. 2501-2505. [cited by applicant]