System and method for detecting fraudulent documents

Patent metadata

Field	Value
Publication number	US-10839208-B2
Application number	US-201816215159-A
Country	US
Kind code	B2
Filing date	Dec 10, 2018
Priority date	Dec 10, 2018
Publication date	Nov 17, 2020
Grant date	Nov 17, 2020

Abstract

A system and method to detect fraudulent documents is disclosed. The system uses a generative adversarial network to generate synthetic document data including new fraud patterns. The synthetic document data is used to train a fraud classifier to detect potentially fraudulent documents as part of a document validation workflow. The method includes extracting document features from sample data corresponding to target regions of the documents, such as logo regions and watermark regions. The method may include updating a cost function of the generators to reduce the tendency of the system to generate repeated fraud patterns.

First claim

We claim:

1. A method of document fraud detection, the method comprising the steps of:
generating a set of synthetic training documents, wherein generating the set of synthetic training documents comprises:
receiving an initial set of training documents;
extracting a document feature from the training documents; and
using a generative adversarial network to generate the set of synthetic training documents, wherein the generative adversarial network receives information about the document feature as input, wherein the generative adversarial network comprises a generator and a discriminator, and the generator comprises a neural network with a generator cost function;

identifying a fraud pattern associated with the set of synthetic training documents;
updating the generator cost function; and
training a document fraud detection system using the set of synthetic training documents.

2. The method according to claim 1 , the method further comprising steps of:
receiving a document;
using the document fraud detection system to detect tampered regions in the document; and
providing an alert that tampered regions have been detected in the document.

3. The method according to claim 1 , wherein extracting the document feature includes using a conditional generative adversarial network.

4. The method according to claim 1 , wherein the document fraud detection system includes a semi-supervised learning algorithm, and wherein training the document fraud detection system comprises training the semi-supervised learning algorithm.

5. The method according to claim 1 , wherein the document feature is associated with a region of a document.

6. The method according to claim 5 , wherein the region is associated with a logo in the document.

7. The method according to claim 5 , wherein the region is associated with a watermark in the document.

8. The method according to claim 5 , wherein the region is associated with a signature in the document.

9. The method according to claim 1 , further comprising:
penalizing the generator for producing repeated fraud patterns in the set of synthetic training documents.

10. The method according to claim 1 , wherein updating the generator cost function further comprises updating the generator cost function to reduce the likelihood that the generator will produce another synthetic document with the identified fraud pattern.

11. A non-transitory computer-readable medium storing software comprising instructions that are executable by one or more device processors to detect fraudulent documents by:
generating a set of synthetic training documents, wherein generating the set of synthetic training documents comprises:
receiving an initial set of training documents;
extracting a document feature from the training documents;
using a generative adversarial network to generate the set of synthetic training documents, wherein the generative adversarial network receives information about the document feature as input, wherein the generative adversarial network comprises a generator and a discriminator and the generator comprises a neural network with a generator cost function;

identifying a fraud pattern associated with the set of synthetic training documents;
updating the generator cost function; and
training a document fraud detection system using the set of synthetic training documents.

12. The non-transitory computer-readable medium according to claim 11 , wherein extracting the document feature includes using a conditional generative adversarial network.

13. The non-transitory computer-readable medium according to claim 11 , wherein the document fraud detection system includes a semi-supervised learning algorithm, and wherein training the document fraud detection system comprises training the semi-supervised learning algorithm.

14. The non-transitory computer-readable medium according to claim 11 , wherein the document feature is associated with a region of a document.

15. The non-transitory computer-readable medium according to claim 11 , wherein the region is associated with a logo in the document.

16. The non-transitory computer-readable medium according to claim 11 , wherein the region is associated with a watermark in the document.

17. The non-transitory computer-readable medium according to claim 11 , wherein the region is associated with a signature in the document.

18. A system for detecting fraudulent documents, the system comprising:
a device processor; and
a non-transitory computer readable medium storing instructions that are executable by the device processor to:
generate a set of synthetic training documents by:
receiving an initial set of training documents;
extracting a document feature from the training documents; and
using a generator of a generative adversarial network to generate the set of synthetic training documents, wherein the generator receives information about the document feature as input;

train a document fraud detection system using the set of synthetic training documents;
identify a fraud pattern associated with the set of synthetic training documents; and
penalize the generator for producing repeated fraud patterns in the set of synthetic training documents.

19. The system according to claim 18 , wherein the document feature is associated with a region of a document.

20. The system according to claim 18 , wherein the generative adversarial network comprises a discriminator, wherein the generator comprises a neural network with a generator cost function, and wherein the non-transitory computer readable medium storing instructions are also executable by the device processor to:
update the generator cost function.

Assignees

• Accenture Global Solutions Ltd

Inventors

• Anthony Samy Julian Addison
• Mannar Kamal
• Le Thanh Hai
• Lim Tau Herng

Classifications

• G06V30/40

  Document-oriented image-based pattern recognition · CPC title

• G06V30/19173

  Classification techniques · CPC title

• G06V10/82

  using neural networks · CPC title

• G06N3/08 Primary

  Learning methods · CPC title

• G06V30/418

  Document matching, e.g. of document images · CPC title